Changes.txt (DC, 03.22.02) -=---------------=- TINI Firmware 1.02e -=---------------=- -------------------------------------------------------------------- Go to http://www.ibutton.com/TINI/book.html for an online version of "The TINI Specification and Developer's Guide". -------------------------------------------------------------------- See API_Changes.txt for changes to the com.dalsemi API. See 1-Wire_Changes.txt for changes to the com.dalsemi.onewire (formerly com.ibutton) API. See Limitations.txt for a partial list of current limitations. See Versions.txt for version information on software that is included in this distribution. Significant Fixes & Changes -=========================- 1.02e -===- -=============- Problem : Calling disconnect() on an HTTPURLConnection can sometimes cause a NullPointerException. Since : Beta Description: If an error occurred in initialization, or disconnect was called before the URLConnection was completely set up, there would be some uninitialized objects, which disconnect would try to use. Solution : Check for null in the disconnect() method. -=============- Problem : Calling Vector.setSize() makes the vector unusable. Since : Beta Description: When calling setSize to make the vector smaller (usually calling setSize(0)) the other elements in the vector should be set to 'null', but instead we accidentally were setting the whole object array to null (forgot to type the array index '[i]'). Solution : Make sure to index into the array. -=============- Problem : Serial Port cannot 'flush()' Since : 1.02c Description: Prior to 1.02c, the serial port always flushed all data before returning control to the system. After 1.02c, TINI would return control immediatly if the serial port was free and the write was less than 1024 bytes. However, there was no way to perform a "flush" operation to wait for the serial port to finish. Solution : SerialOutputStream.flush() now waits for the serial driver to complete all queued operations before returning control to the system. -=============- Problem : Transferring large amounts of data (TCP) over PPP is unreliable at low speeds Since : 1.0 Description: The network stack could overrun the PPP transmitter, resulting in dropped packets. At slow speeds, the roundtrip times were larger than the TCP timeouts resulting in excessive retransmissions. Solution : Added flow control to TCP, added configurable TCP timeouts and segment sizes. -=============- Problem : TINI ignores data received after an HTTP post command. Since : 1.0 Description: When TINI receives a FIN packet, it immediately shuts down the connection without waiting for data. Certain servers send FIN packets with a high sequence number before sending the actual data with a low sequence number (perfectly legal TCP). Solution : Modify TCP to ignore out-of-sequence FIN packets. -=============- Problem : TINI doesn't get a DHCP address. Since : 1.0 Description: TINI used a 3 second timeout waiting for DHCP replies. A new version of a popular DHCP server takes 3.2 to 5 seconds to reply. Solution : Increase the timeouts to 8 seconds. -=============- Problem : TINI's FTP server doesn't work well with CuteFTP's 'UP directory' button. Since : Beta Description: CuteFTP uses the optional 'CDUP' command, which is the equivalent of issuing a 'CD ..' command. Solution : Implement the 'CDUP' command in the FTP server. -=============- Problem : Javakit would still let certain files overwrite bank 0. Since : Beta Description: A code path was exposed that would allow certain files to freely overwrite bank 0 without even asking first. This is very rude behavior. Solution : Add more bank 0 protection in JavaKit. -=============- Problem : Slush's DateCommand would incorrectly account for time zone offsets. Since : 1.02d Description: If you typed in '123000 +6' (i.e. 12:30 at time zone offset +6 hours) your time would come out as 6:30. Solution : Problem fixed. -=============- Problem : Constant strings would sometimes get swept away by the GC. Since : Beta Description: When creating a new string from a constant pool string during periods of heavy memory use (a la TINIHttpServer) it was possible for the new string to be collected before it was returned. Solution : Don't invoke GC on String creates. -=============- Problem : Multi-dimension array dimensions sometimes incorrect. Since : Beta Description: A 'obj = new int[3][2][1]' would think that the object at obj[0] was an 'int[][]', but obj[1] was an 'int[]', and obj[2] was an 'int'. Solution : Fixed the dimension issue. -=============- Problem : EchoCommand did not generate a newline. Since : Beta Description: N/A Solution : EchoCommand generates a newline. -=============- Problem : HTTP operations would sometimes incorrectly eat information it thought was a header. Since : Beta Description: The HTTP classes assumed that all HTTP operations would receive a header. Solution : Check the first few bytes of output to be "HTTP/1" before assuming that there might be a header. -=============- Problem : November 30, 2001 DST behavior incorrect. Since : Beta Description: Certain last days of the month would exhibit incorrect DST behavior due to bad results from Clock.getDayOfWeek(). Solution : Fix the method Clock.getDayOfWeek(). -=============- Problem : Long 'to' strings for the mailto protocol would get chopped off. Since : Beta Description: The write method in the mailto protocol would skip bytes if the buffer was full due to an indexing problem. Solution : Fix the indexing problem in 'MailToPrintStream'. -=============- Problem : A Thread stopped with Thread.stop() could lock the TINI up. Since : Beta Description: A Thread stopped by Thread.stop() doesn't get a chance to release its locks. Solution : Implement a function to release all the locks held by a particular thread. -=============- Problem : Calling read() on a closed InputStreamReader caused a NullPointerException. Since : Beta Description: The underlying input stream was set to null to denote the stream as closed. Solution : Check the underlying stream for null and throw an IOException if it is null. -=============- Problem : Object.wait(long, int) did not check for illegal nano values. Since : Beta Description: The nanosecond argument to Object.wait() must be between 0 and 1000000. Solution : Check for incorrect nanosecond values. -=============- Problem : Hashtable.put() returned incorrect values when replacing an object in the hashtable. Since : Beta Description: put() would return an instance of the private inner class HashtableEntry, rather than the actual Object that HashtableEntry encapsulated. Solution : Return the correct object. -=============- Problem : Vector.indexOf() and lastIndexOf() throw NullPointerExceptions. Since : Beta Description: When a null object is inserted into the list, these methods would not handle their searches correctly. Solution : Handle case when a null might exist in the Vector's list. -=============- Problem : When starting processes in the background, you could get NullPointerExceptions when reading from standard input. Since : Beta Description: When starting in the background, SystemInputStream's root stream was set to null. Solution : If the root stream is set to null, change it to a NullInputStream. -=============- Problem : Telnet through Windows XP echoes everything twice. Since : Beta Description: Although all other Telnet Clients assume that the server will echo, Windows XP assumes that it needs to do the echoing. This is a problem on TINI because TINI does no option negotiation automatically, unlike other servers. Solution : Change TINI to immediately request that TINI does the echoing. -=============- Problem : com.dalsemi.tininet.http.HTTPServer.serviceRequests can leak sockets if a failure occurs spawning a new Thread. Since : Beta Description: serviceRequests expects the new Thread to clean up the socket when it is finished using it. If the Thread is not created the socket is never closed. Solution : Added a catch Throwable block to clean up the socket when any exception is thrown in serviceRequests. -=============- Problem : Reading the real time clock turns off interrupts for an unacceptable period of time. Since : Alpha Description: N/A Solution : Time with interrupts turned off was reduced to 86 microseconds -=============- Problem : No version control for native libraries (.tlib). Since : Beta Description: Running code in a tlib file built for a different firmware revision can cause random code execution and could ultimately cause damage to memory structures and files. Solution : Added support in TINIOS/a390 for checking/inserting hardware and firmware revision information. System.loadLibrary will now throw an UnsatisfiedLinkError if it finds the version information that does not match the current version. Legacy tlib files will still load but cannot be checked against the hardware or firmware revisions. See Native_Methods.txt for more information. -=============- Problem : No support in native library source for NESTED conditional blocks. Since : Beta Description: N/A Solution : Added support in macro(.exe) for nested conditional blocks. -=============- Problem : No support in native library source for $include statements within conditional blocks. Since : Beta Description: Macro, the native library macro and equate preprocessor, was recently updated to support conditional blocks but only assembler code was allowed in these blocks. Example: DEBUG_ON EQU 1 ;... IF (DEBUG_ON != 0) ; spew some debug using info_* and debug_* functions ENDIF Equates and macros were always processed regardless of whether they were in conditional blocks or not. Solution : Macro now supports $include statements within conditional blocks. Example: DS_400 EQU 0 ; IF (DS_400 != 0) $include(ds400.inc) ENDIF IF (DS_400 == 0) $include(ds390.inc) ENDIF This will allow the use of different macro and equate definitions based on a conditional statement. -=============- Problem : Using TINIExternalAdapter changes the external interrupt trigger to "level trigger" Since : Alpha Description: The problem did not allow the use of the DS2480 based 1-Wire and edge triggered external interrupts. Solution : Fixed. -=============- Problem : CanBus Media ID Mask Enable does not work. Since : Alpha Description: All the CanBus methods dealing with Media IDs did not work correctly. Solution : Fixed. -=============- Problem : Domain name picked up by DHCP contains extra trailing NULL character preventing DNS from working. Since : 1.0 Description: Some DHCP servers (incorrectly) append a NULL character. Solution : Modify DHCP to ignore the trailing NULL. -=============- Problem : Native API function System_InstallInterrupt could return an invalid pointer. Since : 1.02 Description: Interrupts that were unused by TINIOS were implemented as a simple RETI statement in the soft IVT. System_InstallInterrupt would assume the implementation of each interrupt in the soft IVT was an LJMP to an ISR. Interrupts that were implented as a RETI instead of an LJMP would incorrectly return 32ffffh as the address. These three bytes were actually the machine code for the RETI instruction itself (032h) followed by two fill bytes (0ffh). If a native library attempted to reinstall the original ISR without disabling the interrupt flag, random code execution would occur and the processor would go into an unknown state, probably resulting in a reboot or other bad behavior. Solution : Changed all Interrupt Vectors to perform LJMPs. Unused vectors LJMP to a RETI statement. -=============- 1.02d -===- -=============- Problem : Overriding ThreadGroup.uncaughtException() did not work. Since : Beta Description: In the VM, we were specifically jumping into the code for ThreadGroup.uncaughtException(), which is basically an empty method. Solution : Rather than point to the code for the ThreadGroup class, get the ThreadGroup object from the VM and point to its class, which makes up point to the correct code. Then added some logic to detect if a thread has entered the uncaughtException method, so that if uncaughtException() throws an uncaught exception, we are not stuck forever in an infinite loop. -=============- Problem : A Thread.read(...) cannot be interrupted Since : Beta Description: There is no way to stop an blocking serial port read. Solution : This is undefined behavior in javax.comm, but the Sun JVM's on Solaris and Windows allow a Thread.interrupt() to interrupt a blocked read without an exception, so we implemented this behavior. -=============- Problem : A thread blocked in CanBus.receive() could block even when the CAN bus was closed. Since : Beta Description: N/A Solution : Any CanBus threads blocked on receive() or sendframe() will be woken up and an exception will be thrown if close() is called on the CAN bus controller. -=============- Problem : SerialPort addEventListener not calling events. Since : 1.0 Description: In javax.comm.SerialPort, The sequence addEventListener( ) notifyOnDataAvailable( ) would work, but notifyOnDataAvailable( ) addEventListener( ) would not, because the event thread would not start. Solution : Check if notify's are enabled in addEventListener, and start the event thread if they are. -=============- Problem : File.lastModified() for a file that does not exist does not return 0. Since : Beta Description: N/A Solution : Fixed. -=============- Problem : Setting the clock with Clock.setTickCount(millis) causes Slush to incorrectly calculate the day of the week. Since : Beta Description: Calling Clock().setTickCount(1002641709L) and issuing a date command in slush results in "Wed Oct 9 15:35:12 GMT 2001". The day of the week should be Tuesday. This was caused by the Date class and the Clock class were using two different techniques to compute the day of the week. Solution : Both classes now use the same code for computing the day of the week -=============- Problem : Slush's 'date' command did not work as advertised. Since : 1.02 Description: Slush's 'date' command should be able to handle setting the time zone to a String like "+10", meaning it is set to GMT + 1000. However, this case was throwing an exception. Solution : Added logic to detect setting a time zone to a number of hours offset from GMT correctly. -=============- Problem : String.indexOf and lastIndexOf returning incorrect values in special cases. Since : Alpha Description: Special cases, such as searching for the empty string starting from a specified location, returned wrong values. For instance... "abc".indexOf("", -99999); was returning '-99999' instead of '0'. Solution : Fixed the special cases. -=============- Problem : Class casting in the virtual machine and java.lang.Class was broken. Since : Alpha Description: Class casting was broken with respect to array dimensions, since we were not checking them. Casting a String[] to a String was an allowed operation. Solution : Begin checking array dimensions when casting. -=============- Problem : URL creation with contexts did not return correct values. Since : Beta Description: Creating a URL in the context of another URL was causing problems, mostly with anchors. If the original (context) URL defined an anchor, and the new URL just defined a file, we were keeping the old context around. Solution : Get rid of the context on creating new URLs in context. -=============- Problem : Class.getSuperclass() was not returning null if the java.lang.Class object represented an interface class. Since : Beta Description: Interfaces don't really have superclasses, even though you can clearly state in the source file that 'interface A extends B', although really in the java class file, A 'implements' B, sort of. Solution : Check for interface class and return null from getSuperclass(). -=============- Problem : We were always truncating milliseconds in the GregorianCalendar class. Since : Beta Description: When calculating the 'long' time value, we were just totally leaving out the milliseconds. Solution : Incorporate the milliseconds. This required several changes to the GregorianCalendar constructors and to Date.parse(String), which were inadvertently relying on bad behavior from the GregorianCalendar calculate methods. -=============- Problem : DatagramSocket.receive() uses memory. Since : 1.0 Description: This is not a bug in TINI. To be compatible with Sun's JDK, TINI has to new an InetAddress object (p.address) on every receive(p). Solution : Add com.dalsemi.tininet.TINIDatagramSocket, a subclass of java.net.DatagramSocket, which overwrites p.address instead of creating a new object. TINIDatagramSocket is a drop-in replacement for DatagramSocket. -=============- Problem : TINI throws an IOException when joining the 224.0.0.1 multicast group. Since : 1.0 Description: 224.0.0.1 is the "all hosts" group, it is not necessary to join this group. However, Sun's JDK accepts this without error. Solution : Change behavior to match Sun's JDK and don't return an error. -=============- Problem : Some outgoing CanBus Frames could be dropped. Since : Beta Description: When filling the outgoing queue, some messages may be dropped instead of being deferred until later. Solution : Fixed. -=============- Problem : Some failed mail connections stay in the connection table. Since : 1.0 Description: If the "." which signifies the end of data fails, an exception is thrown without properly closing the socket. Solution : Close the socket when "." fails. -=============- Problem : java.lang.Thread interrupt behavior did not match behavior of Sun's JDK 1.1 implementation. Specifically, Sun's implementation registered interrupt() calls even when a Thread was not sleeping or waiting. Calls to interrupted() and isInterrupted() returned true in this case. Also, when/if a Thread actually did call sleep, wait, etc., sometime after a call to interrupt, an InterruptedException would be thrown. Previous versions of TINIOS registered interrupts only if interrupt was called on a Thread that was sleeping, waiting, etc. at that moment in time. Since : Beta Description: Behavior of interrupt is poorly documented in javadocs and JLS. Solution : TINI's interrupt behavior was changed to match Sun's JDK 1.1 behavior. -=============- Problem : CanBus IO could be faster. Since : Alpha Description: Much of the object and field parsing of a CanFrame is done in java before a send, or after a receive. Solution : All parsing/building moved into native with over 40% increase in performance. -=============- Problem : Macro (native library macro preprocessor) inadvertantly commented out valid assembly statements in a MACRO. Since : Beta 3 Description: A MACRO with a comment on the same line as the "MACRO" keyword would transfer the comment to the next line. If there was an assembler statement on this line the statement would be commented out. Solution : Fixed. -=============- Problem : TINI throws an exception when writing 0 bytes to a closed socket. Since : 1.0 Description: Sun's reference JDK writes 0 bytes to a closed socket without throwing an exception. The behavior of writing 0 bytes is not specified. Solution : Change behavior to match reference JDK -=============- Problem : Timeouts on serial ports would not function properly. Since : 1.02c Description: A change in the Serial Layer had a misplaced timeout argument. Solution : Fixed the argument. -=============- Problem : Thresholds on serial ports did not default properly. Since : 1.02c Description: The javax.comm default for serial input is to have a 1 byte threshhold. This was accidentally turned off in the new version, making all array reads block by default Solution : Fixed the threshhold. -=============- Problem : Internal Serial Port event misfires and lost events. Since : 1.0 Description: A couple of bugs... because the bitmasks of serial port events matched each other, it was possible for the 'data available' event to be interpreted as a 'parity error' by the TINI Serial Driver, which would send the parity error notification. This probably would not happen on serial port 1, though, because a missing 'ajmp' would make it possible for a parity error to be mis-interpreted as a Carrier Detect, and fire that event instead (even if it was not enabled), losing the parity error in the process. Solution : Fixed event handler code. -=============- Problem : PPP interface not removed on reboot Since : 1.0 Description: The addInterface did not correctly set the ephemeral interface flag and the reboot code did not reset the number of interfaces Solution : Corrected both problems. -=============- Problem : DHCP issues Since : 1.0 Description: TINI would have problems with multiple DHCP servers on the same network, especially after a reboot. Solution : Use unique transaction IDs to identify the servers. Extract the IP from every DHCP ACK packet and make sure that the TINI's IP is cleared before starting DHCP. -=============- Problem : Serial Port notification threads are not released on serial port close. Since : 1.0 Description: The "stopEventThread" native method, though a series of cascading errors, would try to wake up a nonexistant thread instead of the event handling thread. Since the event handler thread would never wake up, it would never terminate (creating a zombie thread). This resulted in the error "java.lang.RuntimeException: Too Many Threads" if the sum of zombie threads + those running normally in the application equals 16. Solution : Fixed. -=============- Problem : DHCP thread died Since : 1.0 Description: The DHCP client did not properly support an SMTP server option that returned more than one IP address. Solution : Fixed. -=============- Problem : When attempting TCP connections, TINI would not retry the first packet. A connection might therefore not succeed. Also, TINI would not include the TCP MSS option when retrying SYN ACK packets. Since : 1.0 Description: TINI did not support SYN retry. Solution : Add SYN retry and the MSS option to the SYN ACK retry code. -=============- Problem : ICMP unreachable messages in response to UDP packets might contain 20 extra bytes at the end, invalidating the ICMP checksum. Since : 1.0 Description: The extra bytes were included in the packet, but not in the checksum calculation. If they were not 0, the checksum would be invalid. Solution : Set correct packet length under all circumstances. -=============- Problem : Some hosts ignore RST packets sent by TINI, sometimes causing connections to stay open on the host side for an extended amount of time. Since : 1.0 Description: Some hosts verify RST packets against sequence/acknowledgement numbers and ignore them if they do not match. The TINI did not always set all fields correctly. Solution : Set correct TCP header fields for all RST packets. -=============- Problem : All mail headers after first header become content Since : 1.0 Description: Certain Windows software interprets end of line characters differently. Solution : Change println() to only write 0x0a. -=============- Problem : There's no way to shut down a mail connection when getOutputStream() throws an exception Since : 1.0 Description: closeConnection() is not public Solution : Make closeConnection() public. -=============- Problem : Sender / Recipient of messages are not recognized Since : 1.0 Description: Certain servers expect <> around SMTP addresses Solution : Add <> in RCPT TO: and MAIL FROM: -=============- Problem : DHCP did not set the mailhost Since : 1.0 Description: TININet.setoptions() did not correctly convert the mailhost IP to a string. Solution : Use the correct IP to string conversion function. -=============- Problem : Malformed packets could stop the telnet and ftp servers. Since : 1.0 Description: The Java code did not handle exceptions thrown by setSocketTimeout(). Solution : Modify exception handling in the telnet and ftp servers. -=============- 1.02c -===- -=============- Problem : TCP initial sequence numbers are predictable. Since : 1.0 Description: This is not a bug, but knowledge of the initial sequence number makes it easier to maliciously take over a connection. Solution : Use truly random ISNs. -=============- Problem : TINI replies to bad TCP flag combinations. Since : 1.0 Description: This is not a bug, but these combinations can be ignored. Solution : Ignore certain TCP flag combinations. -=============- Problem : External Serial Port Lockup Since : Beta Description: Serial 3 lockup was being caused by the 16550 reporting CTS being asserted when flow control was disabled. This caused the write to incorrectly abort, which would cause the IO subsystem to go into an infinite wait. This chain of events have only been observed with serial3 writing while serial2 reads at 115200 Solution : Added an extra check in the external serial port drivers to ignore the 16550 if flow control is not enabled. -=============- Problem : getHostname() and getDomainname() do not reflect changes made by other processes Since : 1.0 Description: Hostname and Domainname are stored in the .tininet file which is updated by setHostname()/setDomainname(). There's also a process-local copy of domainname and hostname which could get out of sync with the .tininet file. Solution : Always read .tininet. -=============- Problem : Serial input and output latency on small (< 10 byte) packets was too high for effective processing. Since : Beta 2 Description: The Serial Drivers would put the process to sleep, even if the data was already available. Also, the Java layer of the serial driver had developed quite large over time, with multiple layers calling into pre-javax.comm code. Solution : The native serial drivers were tweaked, and the Java layer was completly eliminated. On small reads, the latency has been cut by more than half. -=============- Problem : Closing a telnet connection abruptly sometimes caused sessions to hang and spew exceptions. Since : Beta Description: SystemInputStream was sometimes swallowing a read of a '-1' from a stream. Solution : Throw an IOException or return a '-1' (as appropriate) from SystemInputStream when it detects a closed stream. -=============- Problem : Some java opcodes are not optimized. Since : Alpha Description: lushr,lshr,lshl,lneg,ladd,lsub,land,lor,lxor, iadd,isub,iand,ior,ixor were not optimized. Solution : Optimized above opcodes. -=============- Problem : It was possible to "reconnect" to dead sockets Since : 1.0 Description: When the death timer hits, a connection is closed on TINI. However, this was not signalled to the application, so there was still a connection number associated with the socket. When a new connection is opened, the connection number becomes valid again and the application starts using the new connection. Solution : Signal the socket layer. -=============- Problem : BufferedReader would omit a single character under certain situations using the readLine() method. Since : Beta Description: When the BufferedReader refills its buffer, if the last buffer ended in the middle of a newline ("\r\n"), its end-index for the next buffer read would be off-by-1, resulting in the omission of one character. Solution : Fix the indexing problem in readLine(). -=============- Problem : External serial port resources for serial3 would not get cleaned up when the process died if both serial2 and serial3 were opened in the same application. This would manifest itself as a PortInUseException thrown on serial3 if the application was run again without rebooting TINIOS. Since : Beta Description: N/A Solution : The offending driver code was fixed. -=============- Problem : System.arraycopy did not throw exceptions with invalid data copies from an Object array to an array of another object type. You could do this... Object[] source = { "1","2", new Integer(3), "4", "5"}; String[] dest = new String[5]; System.arraycopy(source,0,dest,0,5); Since : 1.01 Description: We allowed any reference array to copy into any other type of reference array. Solution : The native array copy now throws an exception if the reference array types are different (and the source type isn't a subclass of the dest type). In System.arraycopy, we detect this difference in array type and perform a slow copy instead of a fast native copy, allowing the VM to detect casting issues. Note this only happens in cases where you try to copy from Object[] to String[]. Copying from String[] to Object[] is still fast. -=============- Problem : ExternalInterrupt based applications (and any native library that used System_RegisterProcessDestroyFunction) could have resources freed by their process destroy functions when another, unrelated process exited. Since : Beta Description: The process destroy callback function was being called each time any process exited. Solution : Only call the process destroy function when the process that registered it exits. -=============- Problem : Alternating calls to SystemInputStream read(byte[]) and readLine() caused an IndexOutOfBoundsException. Since : 1.01 Description: The internal count of number of bytes available was being decremented by the wrong amount. Solution : Correct the number of available bytes logic. -=============- Problem : In raw mode, SystemInputStream.read(byte[]) did not block as advertised. Since : 1.01 Description: In raw mode, we were checking to see if any data was available before going down to the underlying stream. Solution : Remove the check for available data. -=============- Problem : Users could not extend Telnet and FTP Sessions to suit their own purposes. Since : Beta Description: TelnetSession, FTPSession, and SerialSession were all final classes. Solution : Removed final tag so user applications can extend. -=============- Problem : If BufferedReader's internal buffer ran out on the '\r' of a line, the following '\n' was reported as a different, empty line. Since : Beta Description: We did not handle the buffer running out correctly. Solution : Fixed. -=============- Problem : People were using the "Update Loader" function in JavaKit when not necessary. Since : Alpha Description: Updating the loader is a very dangerous operation, because its possible to get a loader-less TINI. Solution : Add a command line switch that must be called if someone wants to update their loader. -=============- Problem : People with bigger than 512K flash needed some way to load their stuff onto TINI. Since : Alpha Description: JavaKit had built in 'protection' against loading into banks other than 1-7. Solution : Added two options to allow loading to banks higher than 7 and to allow for different flash sizes. -=============- Problem : BuildDependency did not report error values usable by Ant and other third-party build tools. Since : 1.02 Description: BuildDependency was catching Exceptions and then simply returning. Callers could not know if the build process failed or succeeded. Solution : Rethrow caught exceptions to indicate failure. -=============- Problem : Some java.io classes are inefficient with memory. Since : Alpha Description: InputStreamReader, OutputStreamWriter, PrintStream, DataOutputStream, and com.dalsemi.shell.server.SystemPrintStream Solution : Rewrote above classes for memory efficiency. -=============- Problem : Applications starting from .startup during SLUSH boot would not know if an IP was acquired when using DHCP. Since : Beta Description: N/A Solution : A small addition was made to slush to wait for the DHCPListener thread to lease an IP then continue on with slush boot. The default wait timeout is 10 seconds, which is changeable in the slush source by modifying DHCPLOCKTIMEOUT. -=============- Problem : DHCP sets only one DNS server IP. Since : 1.0 Description: DHCP sets the same DNS server IP for both primary and secondary DNS server. Solution : Fixed array source index. -=============- Problem : I2CPort performed an illegal start/stop condition. Since : 1.02 Description: N/A Solution : Fixed. -=============- Problem : I2CPort did not allow for SCL wait states by slave. Since : 1.02 Description: Slow I2C slaves were not able to keep up with a constant data rate, and held SCL low. The I2C driver now detects this condition and slows the bit rate to match the slave. Solution : Fixed. -=============- Problem : Reopening a CAN controller would sometimes lock up the process. Since : 1.0 Description: The order of operations on CAN controller initialization would leave the controller in a lock up state in some situations. Killing and restarting the process would not clear the problem. A reset was necessary to reset the CAN controller hardware. Solution : Fixed. -=============- Problem : Thread.interrupt interrupts a sleeping Thread but does not throw an InterruptedException. Since : 1.02b Description: Internal state was not being marked as interrupted. Solution : Mark state as interrupted. -=============- Problem : Multicast packets fail after changing the IP address. Since : 1.0 Description: The Multicast Java class had a static initializer which copied the IP address. Solution : Change the initializer so it runs every time a new object is created, updating the IP address. -=============- Problem : Repeated reload in Internet Explorer would kill random connections on the TINI or even crash it. Since : 1.0 Description: Handling of RST packets could cause destruction of socket data structures under high load or slow connections or slow server software. Solution : Fixes to the nettcp module. -=============- Problem : Write permissions not checked. Since : Beta Description: When writing over or appending to a file that already exists, the write permissions for that file are ignored. Solution : Firmware changed to check for write access. -=============- Problem : TINI's DHCP client would not renew its IP address Since : 1.0 Description: Several issues in the DHCP client, among them: - sends packets with source IP address, although no IP has been leased - all packets contain invalid fields, fields are not properly filled in for state - sends unsolicited address request package after each IP release, invalidating the release packet - never enters rebinding state - loses IP address after 1/2 of lease time, cannot renew leases if a DHCP relay is used - does not stay in renewing state for proper amount of time - does not properly notify of IP loss in all cases - does not invalidate IP if IP is lost Solution : Fixes to Slush and DHCPClient. -=============- Problem : Multiple TINIs on the same network would generate an unusual amount of IGMP messages on the network if all of the TINIs belonged to the same group. Since : 1.0 Description: When receiving an IGMP report message, TINI would transition to the DELAYING state, thus ultimately generating a report. Solution : Conform to RFC1112 and transition to IDLE state instead. -=============- 1.02b -===- -=============- Problem : Receiving all TCP data from TINI sometimes took a long time. Since : Alpha Description: When a TINI application closed a TCP socket which still had data in its output buffer, the firmware cleared the HAVE_OUTPUT_DATA bit. The data would then only be sent at the next retry. Solution : Don't clear that bit. -=============- Problem : TINI would try to communicate on the network even if it didn't have a MAC address part. Since : Alpha Description: N/A Solution : Disable Ethernet if MAC ID is zero. -=============- Problem : LineNumberReader would throw NullPointerException if the underlying stream threw an exception. Since : 1.02 Description: Actually this was a more general problem with exception block and Java method boundaries. It showed up in LineNumberReader under certain circumstances. Solution : Fixed. -=============- Problem : TINI was unable to send mail to certain mail servers. Since : 1.0 Description: mailto protocol class did not properly handle multiline server replies as specified in RFC 821 appendix E. Solution : Added handling of multiline replies. -=============- Problem : Integer/Long decode() did not correctly report a bogus digit, but would always print "-1" Since : Alpha Description: Decimal would always be -1 because that is thei error code when we can NOT convert a digit. Solution : Print charAt instead of decimal. -=============- Problem : DHCP Release messages contain garbage options after "end of options" tag. Since : Beta Description: Data buffer was reused and not properly cleared. Solution : Clear end of buffer. -=============- Problem : Error messages like "404 - Not found" generated by HTTPServer would not display correctly. Since : 1.0 Description: arraycopying the error message overwrote the HTML tag. Solution : Increase the buffer index before copying the error message. -=============- Problem : When browsing TINI web pages with Internet Explorer, it would sometimes display parts of the HTTP header, corrupting display output. Since : 1.0 Description: When resending SYN ACK packets due to a timeout, the code would not supply the connection number to the function responsible for increasing the sequence number. The first data packet would therefore have a sequence number too small by one, destroying the HTTP reply header. Solution : Supply the correct connection number. -=============- Problem : gftp ftp client would not work with TINI ftp server. Since : N/A Description: gftp uses LIST -L instead of LIST. Solution : Ignore the -L parameter. -=============- Problem : TINI ftp server didn't implement NOOP, MODE, STRU. Since : Alpha Description: RFC 959 mandates that a minimal RFC compliant ftp server must implement these commands. Solution : Added these commands. -=============- Problem : TINI ftp client broken (list, get, put terminate session). Since : Alpha Description: The ftp client is not multithreaded and waits for a result of a command not yet issued. Solution : Added a work-around which ignores the result. -=============- Problem : TINI sent funny RST packets. Since : Alpha Description: The urgent field in the TCP header was not cleared. Solution : Clear the field before sending RST. -=============- Problem : TINI DNS cache would not timeout and expire entries Since : Alpha Description: DNS entries are cached, and not rechecked for expiration causing possible problems with DNS servers that modify IP address mappings frequently. Solution : Added DNS cache timeout. -=============- Problem : TINI would send ICMP port unreachable packets if it wasn't listening to a port even though the destination address was a multicast address. Since : 1.01 Description: N/A Solution : Added check for multicast address to ICMP function. -=============- Problem : TINI would use up all kernel memory if it received unsolicited ICMP echo responses. Since : Alpha Description: When not expecting ICMP echo replies, the ICMP module just exited without freeing the packet buffer. Solution : Added call to 'free' buffer. -=============- Problem : com.dalsemi.tininet.TININet.getSubnetMask would only work for eth0. Since : Alpha Description: Static string in function call to network module. Solution : Fixed call to use interface parameter instead of static string. -=============- Problem : Network packet flood triggered hardware watchdog. Since : Alpha Description: The CPU was spending all its time in the Ethernet interrupt handler, preventing the execution of the scheduler. Solution : Changed Ethernet interrupt so it leaves network interrupts disabled and have the scheduler reenable them after checking the watchdog. This leaves about 20% of the CPU to do other work than handling Ethernet interrupts. -=============- Problem : Random reboots in network module. Since : Alpha Description: Both the Ethernet interrupt and IOPoll accessed the same queues, but there were no semaphores to control access. Solution : Protected the network queues against simultaneous access. -=============- Problem : Malformed packets would crash TINI. Since : Alpha Description: Buffer overruns occurred because TINI trusted that packet lengths and parameters were truthful and legal. Solution : Added bounds checking to several network modules. -=============- Problem : Properties was not trimming data keys and values. Since : 1.02 Description: Had taken out the call to trim() because tests made it look like Sun's implementation did not do it. Solution : Changed to trim the data keys and values. -=============- Problem : System, Telnet and SerialInputStreams would behave poorly when used by non-Slush applications by throwing exceptions and not handling backspaces correctly. Since : Beta Description: The streams were not designed with non-Slush, non-readLine only uses in mind. Solution : SerialInputStream, TelnetInputStream and SystemInputStream changed to handle general cases so an application could read data with embedded backspaces. -=============- Problem : Properties was locking up on load if someone was using Unicode characters. Since : Beta Description: Properties was draining an exception and never incrementing an index into an array caused an infinite loop. Solution : Fixed to handle UTF8 data. -=============- Problem : Telnet sessions closed under some circumstances would leave idle threads running around, if a process was running in the background. Since : Beta Description: When catching an exception on a SystemInputStream (usually TelnetInputStream), if the stream was part of a background process, it would not end the telnet session. Solution : End the telnet session on all errors. -=============- Problem : When reading a property file (with the Properties class) that only used '\r', incorrect keys would be returned. Since : 1.01 Description: Not handling the case where '\r' ended a line correctly in the DataInputStream class. Solution : Changed to handle this case correctly. -=============- Problem : CAN1 has trouble initializing correctly in some cases. CAN0 works correctly and is unaffected. Since : 1.02 Description: sendFrame() and receive() would throw exceptions due to disabled CAN controller one (CAN1). Controller was not initialized correctly. Solution : Fixed incorrect mask in initialization check. -=============- Problem : Thread.interrupt causes IllegalMonitorStateException on a Thread that is blocked on wait. Since : Beta Description: N/A Solution : Fixed. -=============- Problem : java.util.Properties broken in 1.02. Since : 1.02 Description: Properties was no longer trimming keys and data. Solution : Roll back to previous version. Properties changes were prompted by tests indicating change was needed when it was not. -=============- Problem : File Redirect Echo. Since : Beta Description: If file re-directed input was specified from slush, the output would be echoed to the screen. Solution : Fixed. -=============- Problem : Mailhost set to Loopback. Since : Alpha Description: Trying to clear the mailhost value (TININet.setMailhost("")) would set it to loopback. Solution : Passing null or "" to setMailhost now sets the mailhost to 0.0.0.0. -=============- Problem : Under extremely heavy memory usage heap corruption could occur. Since : 1.02 Description: There is a chance that the garbage collector could corrupt memory. However, normal operation will not reproduce the problem. Under extremely heavy memory allocations, when the garbage collector runs continuously and the user is allocating memory continuously, it is possible to have heap corruption occur after several hours. Solution : Memory corruption issue has been fixed. -=============- Problem : I2C master ACKs last byte on reads from slave. Since : 1.02 Description: The TINI I2C master firmware will ACK the last byte of a read from an I2C slave contrary to the protocol spec. Solution : Removed ACK on last byte of reads. -=============- Problem : SerialInputStream.close was sychronized improperly. You could not call close if another thread was blocking on read. Since : 1.0 Description: N/A Solution : Close method is no longer synchronized waiting for the read to complete. -=============- Problem : Removing the last serial port event listener would not remove the listening thread. Since : 1.0 Description: When an application registers an event listener for the serial port and then removes it, the listening thread would not be removed. This results in a thread leak. Solution : Listening thread is stopped when the event listener is removed. -=============- Problem : Float.equals and Double.equals do not return true when comparing two NaNs that have binary values that are not the same. Since : Beta 1.0 Description: Binary comparison was being performed for all values. Solution : Added extra test for NaN. -=============- Problem : javax.comm.SerialPort.setRcvFifoTrigger throws IllegalArgumentException. Since : 1.02 Description: N/A Solution : Fixed. -=============- Problem : Serial port 0 periodically dropped data. Since : 1.02p2 Description: TINI periodically dropped serial data Solution : Put critical code in protected sections. -=============- 1.02 -==- -=============- Problem : The native methods Mem_Copy*, Mem_Clear* and Mem_Compare* did not set DPS to a known value. This could result in the change of the source and target roles if DPS was not zero when these methods were called. Since : Beta Description: See Native_ReadMeNOW.txt for more information. Solution : Fixed. -=============- Problem : Calling com.dalsemi.system.TINIOS.enableSerialPort1(boolean) alternating true and false would cause a system hang. Since : 1.0 Description: N/A Solution : Stopped serial driver from registering too many times. -=============- Problem : 1-Wire subsystem needs updating to new long line testing results. Since : 1.0 Description: Long line testing in the lab has resulted in new info about 1-Wire and long lines. The driver now uses better parameters for long lines in both Standard Speed and Flex Speed modes. Solution : N/A -=============- Problem : I can run a class that has a "public void main(String[])" instead of a "public static void main(String[])". Since : TINIConvertor Description: To make the application, we just looked for any method "public void main(String[])", and didn't check to see if it was static or not. Solution : Check to see if the method is static or not. -=============- Problem : Datagram Packet did not check buffer length Since : Since TINI crawled out from the primordial goo Description: DatagramPacket did not check the length of the packet array given by the user before sending or recieving packets. If the packet length specified is greater that the array length, this could potentially lead to memory overruns. Solution : DatagramPacket setLength() and setData() now throw an IllegalArgumentException if the packet length is greater than the datagram length. If setData() is called and the existing length is greater than the new array's length, then the length of the datagram will automagically shorten itself to the size of the new datagram array. -=============- Problem : Memory leak in File.delete() and File.renameTo(). Since : Beta Description: N/A Solution : Fixed. -=============- Problem : Enabling serial1 many times causes a reboot Since : 1.02p1 Description: N/A Solution : Fixed enable code to not reboot system. -=============- Problem : Parity on serial1 does not work. Since : 1.02p1 Description: N/A Solution : Send the proper parity bits. -=============- Problem : Several memory leaks fixed on the internal serial ports. Since : 1.02p1 Description: Changing buffer sizes, enabling serial port 1 and closing the serial ports caused memory leaks. Solution : Free the memory when done with it. -=============- Problem : Trouble recycling PPP objects. Since : Beta Description: PPP object was not releasing native resources. Added stopPPPThread method to allow application to kill PPP daemon in event thread. Exposed freeInterfaceWrapper method to allow applications to release native resources. Solution : Call these two functions before removing last reference to PPP object. -=============- Problem : Problem connecting to ISP's Since : Beta Description: PPP HDLC decoder was being confused by HDLC runt packets. Solution : HDLC decoder now ignores runt packets. -=============- Problem : PPP native resource leak. Since : Beta Description: If an application that contained PPP objects was killed native resources were not being freed. Solution : PPP now frees all native resources when application exits or is killed. -=============- Problem : DHCP client not setting domain name after acquiring IP address. Since : 1.02p1 Description: DHCP client does not set domain name. Solution : Fixed DHCP client. -=============- Problem : The getLastModified() method of URLConnection would throw an exception if the URL response contained a valid last-modified header, and the getHeaderFieldDate() throws exceptions on invalid data, rather than returning the default. Since : Beta Description: URLConnection was trying to parse the last modified field as an Integer, and not catching an Exception on parsing in the getHeaderFieldDate() method. Solution : Use the Date.parse() method in getLastModified, and catch the exception and return a default in getHeaderFieldDate(). -=============- Problem : Telnet stability Since : Beta Description: If the accept() call in Telnet throws an IOException due to too many sockets, we need to wait to try to call the accept() method again, or else we will see lots of IOExceptions chewing up TINI's memory. Solution : Implemented an exponential backoff starting at 100 ms and maxing out around 5 minutes in TelnetServer. A successful connection resets the backoff timer. -=============- Problem : Calling Calendar.set(HOUR, x) doesn't correctly set the hour. Since : The birth of the .util package. Description: There are two hour-like field for calendar, HOUR and HOUR_OF_DAY. We were only looking at HOUR_OF_DAY in our internal calculations. Solution : Also look at HOUR, but we also have to determine which was set most recently. Implemented a timeStamp and modified[] table to track when fields are set relative to each other. In GregorianCalendar, changed the computeTime() method to actually use this data. -=============- Problem : BuildDependency's dependency on ';' might not make Solaris users happy. Since : 1.02p2 Description: There was some concern that a semi-colon would be interpreted as a separator on the command-line. Solution : BuildDependency now treats ',' commas exactly as semi-colons. Also, the BuildDependency defaults have been updated to reflect a new OneWireContainer, family code 33. -=============- Problem : Another flavor of Telnet Death Since : Beta Description: The accept() method in TelnetServer would throw exceptions even though netstat only reported two sockets being used. Solution : When sockets were being closed and opened at about the same time, the internal socket count would not decrement because the socket would not validate in some conditions. Added a weaker validate method in the event a socket was closing to make sure there are no 'ghost' sockets. -=============- Problem : Can't load tbin files to banks larger than 7 in JavaKit. Since : Beta Description: We were explicitly checking against that. Also, we were sending the bank number as (bankNum + 0x30) which is bad for any bank higher than 9. Solution : Added an "-allow" option to allow loading to banks higher than 7. Also, correctly implemented sending the bank number to the loader. -=============- Problem : Double toString representation accuracy. Since : Beta Description: I noted that some numbers in certain ranges usually printed out more digits than in other ranges. Solution : Added a heuristic to account for this. -=============- Problem : Date.parse() method chokes if there is something like 'Sat' in the String. Since : Beta Description: Although ignored, Strings like 'Sat', 'Sa', and 'gm' are allowed to be in the Date String, while 'BC' is not. Solution : Added a String of allowed tokens to check against. -=============- Problem : BitSet failed API conformance tests. Since : Beta Description: You could call BitSet's constructor in such a manner that the size of the BitSet was not a multiple of 64. Also, the hashCode() method occasionally returned incorrect values. Solution : Fixed a logic error in the constructor and a sign-extension issue in hashCode. -=============- Problem : GetOpt throws an ArrayIndex Exception after its reported a failure. Since : 1.02p1 Description: There was one spot in the code we were incrementing the pointer without making sure we were still in bounds. This would only occur if there were errors in the arguments, i.e. an argument was expected to follow a switch. Solution : Check the index before accessing the array. -=============- Problem : SimpleTimeZone failed API conformance tests. Since : Beta Description: We had no toString() method, and hashCode produced incorrect results. Also, the clone() method always set useDaylight to true. Solution : Added toString(), fixed hashCode(), and preserved the value of useDaylight. Note that toString() still does not match Sun's implementation, because Sun is for some reason reporting internal implementation details in their toString() method. We implemented differently, so this data does not appear in our toString() result. -=============- Problem : If I get a TimeZone and mess with its rawOffset, then get that TimeZone again later, it still has the skewed rawOffset. Since : Beta Description: We were giving out the VM's only copy of the TimeZone. Solution : Clone the TimeZone before handing it out in TimeZone.getTimeZone(). -=============- Problem : FTP connections through the URL classes don't return the right content-length. Since : 1.02p1 Description: We did away with the FTPInputStream in the interest of speed and API size, although it was storing the content-length for us. Solution : We had the data stored elsewhere by parsing the FTP server's response line (where it says "(54324 bytes)"). -=============- Problem : FTP requests through a proxy didn't work. Since : Beta Description: We have to add a request header field 'Host' to make the URL stuff work with certain flavors of Microsoft Servers that barf if that field is missing. However, FTP proxy's don't like this header field. Solution : Added a method in com.dalsmei.protocol.http.Connection setFTP() to tell the connection that it is an FTP proxy connection, so it knows if it needs to skip cramming the 'Host' field onto the headers. Also, com.dalsemi.protocol.ftp.Handler has to tell the new connection that it is a procy FTP connection. -=============- Problem : Proxy settings aren't handled right. Since : Beta Description: Sun uses system (per-process) properties "ftpProxySet", "ftpProxyHost", and "ftpProxyPort" for FTP proxies, and "http.proxyHost"/"http.proxyPort" OR "proxyHost"/"proxyPort" for HTTP proxies. We were only looking in TININet for system-wide properties. Solution : Change the URL-related classes to search the System properties for proxies. This means that programs that had earlier had lines like: TININet.setProxyHost("1.1.1.1"); TININet.setProxyPort(8088); Will need to change to have lines like this: System.getProperties().put("proxyHost", "1.1.1.1"); System.getProperties().put("proxyPort", "8088"); Depending on which proxy will be used. -=============- Problem : Putting a file in FTP to a directory that did not exist kills the session. Since : Beta Description: An uncaught exception bubbled up and whacked the session when the FileOutputStream was created to the non-existant directory. Solution : Catch the exception and print a nice message about the failure. -=============- Problem : You can't clear out the proxy settings in TININet once they're set. Since : Beta Description: By passing null or "" to setProxyHost, you should be unsetting any proxy settings, but we pass that value to InetAddress, which returns localhost as its default in event of failure. That's no good. Solution : Catch those values and change the input String to "0.0.0.0", which equates to NOT_SET. -=============- Problem : Locale failed API conformance tests. Since : Beta Description: Variant settings were destroyed. We were setting the internal variant to the value of the Country passed in. Solution : Don't do that. -=============- Problem : ListResourceBundle and PropertyResource bundle can report duplicate keys. Since : Beta Description: Happens when a child resource bundle's keys overrides the parents. In this case, the child's key should be returned. Solution : Use a Vector to accumulate the child and parents keys, watching for and ignoring any duplicates. -=============- Problem : getObject() in ResourceBundle always throws a MissingResourceException when the parent has the object. Since : Beta Description: We weren't setting up the parent hierarchy when creating a resource bundle. Then, in getObject(), if we can't find the key in the child, then we look in the parent. But then, irregardless of success or failure, we threw an Exception. Solution : Set up the parent hierarchy, and check for success when getting the object from the parent. -=============- Problem : Date failed API conformance tests. Since : Beta Description: UTC() throws an exception, and toGMTString() does not put a '0' in front of single-digit dates. Solution : The UTC() method was changed to use "GMT" time zone rather than "UTC", which does not exist on TINI. Changed the String formatting code to put the extra '0' in the String when needed. -=============- Problem : In a time zone with DST, setting a Calendar's date to March 100th doesn't change us to Daylight Sacings Time. Since : Beta Description: March 100th==June 8th, so it should be in DST, but we were getting our DST offset before with the month. Solution : If the date is out of valid range (1-31), recalculate the DST offset with the new month and date. -=============- Problem : StreamTokenizer failed API conformance tests. Since : Beta Description: We weren't interpreting C-style comments or NUMBER digits correctly. Solution : For C-style comments, we saw the '/' part of '/*/ and assumed it was a normal comment to the end of the line. Also, if we found a lone '.' in the stream, since it is a NUMBER type, we tried to call Double.valueOf(String) on it, and got an exception, but then decided it was a WORD type, which it wasn't. -=============- Problem : Properties calls trim() on any keys it loads. Since : Beta Description: It should not trim() the keys. Solution : We have decided to implement Properties() to the 1.2 style. See limitations.txt for more information. -=============- Problem : Bloat. Since : Beta Description: Many classes were using for-loops to fill and copy arrays, and not sharing that resulted in lots of duplicate code. Solution : The following classes have been scoured to replace for-loops with methods from the class com.dalsemi.system.ArrayUtils whenever possible, and also to use the newly public IPString methods in TININet (see API Changes): com.dalsemi.tininet.http.HTTPServer com.dalsemi.tininet.dns.DNSClient com.dalsemi.tininet.dhcp.DHCPClient com.dalsemi.tininet.TININet java.net.InetAddress. -=============- Problem : ResourceBundle failed API conformance tests. Since : Beta Description: We weren't throwing an Exception if a bundle could not be found. Also, when searching for a bundle under a given locale, if we could not find it we weren't looking for the bundle under the default locale as well. Solution : Fix the above inconsistencies. -=============- Problem : Java code verifier turned up some inefficiencies in code. Since : N/A Description: Many classes had logic inefficiencies, like a comparison statement that was always the same or a mathematical operation that always had the same result. Solution : The following classes contained such inefficiencies, and they were removed: java.util.Date java.util.GregorianCalendar java.util.SimpleTimeZone java.lang.Math com.dalsemi.tininet.http.HTTPServer java.io.FileInputStream java.io.FileOutputStream java.io.RandomAccessFile. -=============- Problem : Integer.parseInt() needs a smarted overflow indicator for errors. Since : Beta Description: It turns out there's this number (429496730) that when multiplied by 10 as an 'int', doesn't overflow by going negative, but it becomes positive, something like +4. Solution : Check if the intermediate value is less that the current value while parsing the integer. -=============- Problem : java.lang.Class failed API conformance tests. Since : Beta Description: Our forName() allowed for Primitive classes (boolean, int) to be loaded, though that shouldn't be allowed. forName() and getPrimitiveClass() were charing the same cache. Also, cast checking needed serious work--Checking the cast on things like int[][]-->Object[], String[][]-->Object[], and Object[]-->Object would fail. Solution : Make two caches, one for primitive classes and one for non-primitives. Also, fix the cast checking method. -=============- Problem : Long's hashCode method doesn't match Sun's implementation. Since : Beta Description: We were jsut casting the long down to an int. Solution : Changed to XOR the high and low 32 bits together. -=============- Problem : Math failed API conformance tests. Since : Beta Description: ceil() was broken horribly. Many other methods failed on special cases (i.e., one number is NaN), or were casting doubles to int's rather than long's. Solution : Changed the ceil() implementation to look more like floor()'s implementation. Also, the following methods were changed to handle special cases or cast to long's rather than int's: log floor rint round(float) round(double) max min. -=============- Problem : String failed API conformance tests. Since : Beta Description: Constructor String(char[], int, int) wasn't checking the second argument to make sure it wasn't negative, so we'd try to create an array and get a NegativeArraySizeIndex, rather than a StringIndexOutOfBounds. Also, in the encoder Constructor, if offset!=0, the encoding process would fail because we would pass the wrong data to the encoder. In method substring(int), we weren't checking for the index out of bounds. Solution : Fixed. -=============- Problem : If you su in a telnet session then exit, your session freezes. Since : 1.02p1 Description: There was a 'return' after the call to end the session, which is bad if you've su'd because then the session's thread has exited and you can't continue on. Solution : Removed the call to return. -=============- Problem : When you log in to an FTP session, it doesn't change you to your home directory like telnet sessions do. Since : Beta Description: N/A Solution : Changed to log in to your home directory if it exists. -=============- Problem : Sometimes a ServerSocket starts throwing infinite IOException on accept(). Since : Beta Description: If a socket was aborted (a RESET sent) before TINI created a socket object for it, we would still try to clean it up. The problem is we haven't set its connection number yet, that memory was just MALLOC'd so its all 0's. So, we end up aborting whatever socket is at connection 0. If thats a server socket, thats bad--then socket validation fails and you can't accept any new sockets. Solution : If the state of the socket recieving the RESET is SYN_RCVD, don't call SOCK_Abort, just clean everything else up. -=============- Problem : Can't bypass the .startup file by hitting the '5' key. Since : 1.02p1 Description: Slush was using System.in.available() to see if any data was waiting, but that implementation of SystemInputStream reportedincorrect values that was fixed by 1.02p1, so Slush never thought any data was available. Solution : Changed Slush to get the root input stream to find available data. -=============- Problem : netstat prints out slowly. Since : Beta Description: It was byte banging. Solution : Use a BufferedOutputStream. -=============- Problem : TINI stays up for days at a time now (I guess that's not really a problem) Since : Alpha Description: N/A Solution : Added a "Days: " line to the 'stats' command in Slush. -=============- Problem : Trying to set Slush's time zone to an invalid time zone gets an exception. Since : Beta Description: N/A Solution : Catch the exception and print out the usage string. -=============- Problem : 'ps' doesn't report owner names. Since : Beta Description: N/A Solution : Report owner names along with process names. -=============- Problem : If I try to execute a directory (duh) with the slush command 'source', I get an exception. Since : Beta Description: N/A Solution : Print out a nice error message. -=============- Problem : Slush does not release CommPortOwnership listener when serialServer is shut down. Since : 1.02p2 Description: N/A Solution : Fixed. -=============- Problem : Serial Ports leak memory when port is opened. Since : Alpha Description: N/A Solution : Fixed. -=============- Problem : Invoking methods with very large (> 15 single-width) parameters lists causes the system to crash. Since : Alpha Description: N/A Solution : Fixed. -=============- Problem : Only the static initializers with non-constant static fields are executed Since : Alpha Description: N/A Solution : All static initializers are now run regardless of the existence of static fields. -=============- 1.02p3 -=====- -=============- Problem : Process destroy routines destroyed system data. Since : Beta Description: N/A Solution : Fixed. -=============- 1.02p2 -=====- -=============- Problem : If a ServerSocket, on port p, is closed and then reopened (on p) a BindException is thrown if any other sockets bound to p are still active. Even if they are not in the LISTEN state. Since : Alpha Description: None Solution : This is fixed for 1.02. As long as no other TCP connections exist on the same port in the LISTEN state a ServerSocket on that port can be created. -=============- Problem : ThreadGroup.remove not called on Thread creation/start error. Since : Beta Description: This problem could manifest itself as errors in active thread count, etc. The problem usually only occurs when an attempt is made to start more than the maximum number of allowed Threads. Solution : Fixed. -=============- Problem : CanBus.close() would cause the CAN controller to become unresponsive until after a reboot Since : 1.0 Description: Controller was reset inproperly in some circumstances. Solution : Fixed. -=============- Problem : TINIOS.enableSerialPort1(true) would not reenable the DS2480 one wire code when serial 1 was released. Since : 1.02p1 Description: N/A Solution : Fixed. -=============- Problem : CommPortOwnershipListener and related methods did not work properly across processes. Since : Beta Description: N/A Solution : Fixed. -=============- Problem : Loading a properties file with java.util.Properties throws an ArrayIndexOutOfBoundsException. Since : Beta 3 Description: This problem only occurred when a key was not followed by a value. For example "my.property=" on a line by itself would cause the exception to be thrown. Solution : Added a check for this case in Properties.java. -=============- Problem : com.dalsemi.TINIOS.getCurrentUID() always returns 0x80. Since : Beta 3 Description: The default owner for Bank 7 application was 0. Slush required root privileges to execute properly so the method was set to always return an admin UID. Solution : The default owner for the Bank 7 application is now 0x80. Other processes will now query the current shell for their current UID. -=============- Problem : Date parsing of Strings did not match Sun's functionality. Since : Beta Description: Date.parse() would not fail when it was supposed to, and sometimes interpreted a date string incorrectly. Solution : Fixed. -=============- Problem : Some Slush commands would throw an ArrayIndexOutOfBounds exception when incorrect/incomplete arguments were given. Since : 1.02p1 Description: GetOpt would sometimes move on to the next argument without making sure there was a next argument. Solution : Change to always check to make sure that the next argument exists. Also added support for 'instant failure', if GetOpt runs into a character or argument it doesn't know what to do with, if the format string used to create GetOpt starts with a '!', it fails on this argument. Default behaviour is to leave the argument for the application. -=============- Problem : An OutputStreamWriter wrapped around an HTTPOutputStream would not write all of its data. Since : Beta Description: OutputStreamWriter flushes when its buffer is full, but when an HTTPOutputStream is flushed, it also closes. Solution : Changed OutputStreamWriter to just write to the underlying stream, and not flush until told to do so. -=============- Problem : Source would cough up an "ERROR:Unknown Command" on blank lines of scripts. Since : Beta Description: Source did not check for empty lines. Solution : Source now checks for empty lines before trying to execute them. -=============- Problem : FTPSessions in passive mode would leave lingering ServerSockets forever. Since : Beta Description: Since FTPSessions are part of the init process, they never clean up those server sockets, and we weren't closing them after a connection was made. Solution : Close the passive server socket after a connection is made. -=============- Problem : Gregorian Calendars would not set a date correctly in March of a leap year. Since : Beta Description: Gregorian Calendar calculates the number of days in a year before many of its other calculations. We add 1 to the calculated number of days in a year if its a leap year AND its past february, but since Java uses funky 0==January month offsets, we accidentally checked if it was later than March. Solution : Change to compare the month to later than February. -=============- Problem : Gregorian Calendar is slow Since : Beta Description: The main calculations use floats and longs. Solution : Tried to replace these calculations with long/int calculations where testing revealed the calculations would be the same result. -=============- Problem : Lots of code duplication between ChartoByteUTF, DataOutputStream, DataInputStream, and RandomAccessFile Since : Beta Description: DataOutputStream and RandomAccessFile use almost the exact same UTF8 transaltion code, which is only slightly different than the CharToByteUTF transaltion code. Added a method in ChartoByteUTF that allows it to do real-UTF (0 transaltes to 1 byte) or java-modified UTF (0 translates to 2 bytes). Also added a method in ChartoByteUTF that RandomAccessFile and DataOutputStream could share. Also, RandomAccessFile, DataOutputStream, and DataInputStream were doing lots of packing of byte arrays to and from longs, ints, doubles, etc. that is better performed now with methods in the com.dalsemi.system.ArrayUtils class. Solution : Fixed. -=============- Problem : The slush 'Source' command makes it hard to do any infinite loop scripts and stuff. Since : Beta Description: It could use some scripting power. Solution : Added some scripting functionality. The following are supported: 1. Executes a line forever. WHILE TRUE normal-command-line 2. Executes a line a set number of times. FOR normal-command-line 3. Compound statements can be used in place of "normal-command-line". WHILE TRUE { do-something-1 do-something-2 do-something-3 } 4. Command line parameters. java echo.tini $1 $2 $3 $4 $5 Parameters which do not exist will be replaced with "". -=============- Problem : HTTP headers were not being treated as case-insensitive. Since : Beta Description: com.dalsemi.protocol.HeaderManager uses a String.equals() comparison. Solution : It now uses String.equalsIgnoreCase(). -=============- Problem : Telnet death. Since : Beta Description: Some telnet death seems to be related to synchronization, so several methods in Session, Server, TelnetSession, and TelnetServer are now synchronized. Also, make a Session bail out if any java.lang.Throwable bubbles up to the run() loop. Also, sometimes it takes a while for Threads to die from the time they first report an IOException, so added more checks to make sure Telnet is still alive (checking its PrintStream.checkError() method) and capturing errors in the TelnetInputStream. Solution : Fixed. -=============- Problem : Using serial port 1 corrupts data and causes random system behavior. Since : 1.0 Description: Opening and using serial port 1 causes a scrog of system data. Solution : No longer writing to memory I don't own. -=============- Problem : Optional PPP slush command does not reflect changes to PPP API. Since : 1.02p1 Description: Optional PPP slush command does not reflect changes to PPP API. Solution : PPP command source has been updated. -=============- Problem : Transferring large amounts of data over a TCP connection over a slow (or congested) link results in incomplete transfers. Since : Alpha Description: N/A Solution : Fixed. -=============- Problem : Some events on external serial ports caused random behavior of the system. Since : 1.02p1 Description: When an external serial event fired, it might wake up the wrong threads. Solution : Wake up correct thread. -=============- 1.02p1 -=====- -=============- Problem : Native libraries must exist in TINI's file system. Since : Beta 3 Description: Requiring native libraries to exist in the file system, makes recovery from file system damage difficult. Solution : Native libraries (.tlib files) can now be attached to any application (.tini or .tbin files) using TINIConvertor. See Native_Methods.txt for details. -=============- Problem : Slush password hashes were insecure--created by appending the password to the username. Since : Beta Description: This is insecure because username 'root', password 'tini' hashes to the same result as 'roott', 'ini' does. An attacker could guess the password in linear time. Solution : Hash (username + ":" + password). -=============- Problem : The AddUser command in Slush checked to see if a user existed in the password file by using the String.startsWith() method on each line of the password file. Since : Beta Description: This is a problem if a user "user_admin" existed, and you wanted to create a user named "user". The operation would fail. Solution : Parse the password from the file entry, compare using the String.equals() method. -=============- Problem : Time zones reported by the "date -t" method were undescriptive. Since : Beta Description: TINI supports several time zones, but the Date command only reported their names, and none of their characteristics. Solution : Changed the command to also print out time zone offsets, and whether or not the time zone uses Daylight Savings Time. -=============- Problem : BufferedReader would not accept '\r' as an end of line character. Since : Beta Description: In environments where only the '\r' character is reported for the end-of-line, this would cause BufferedReader to wait for another end-of-line character. The data reported by BufferedReader.readLine() would then be 'delayed' by a line. Solution : Change to recognize '\r' as an acceptable end-of-line character. -=============- Problem : Boot state information introduced to the API. Since : N/A Description: N/A Solution : The StatsCommand has been updated to report all boot state information, such as occurrances of heap clear and file system modification. -=============- Problem : The getOffset() method of SimpleTimeZone reported incorrect values. Since : Beta Description: In CST/CDT, TINI thought Thursday, October 19 was not in Daylight Savings time, but the 18th and 20th were. In addition, TINI ended DST one week early (on October 22nd). Further testing revealed more inconsistencies with Sun's implementation. Solution : Fixed it, and then tested against Sun's implementation thoroughly. TimeZones supported by TINI now report the correct offset from GMT. -=============- Problem : SystemInputStream.available() reported incorrect values. Since : Beta Description: SystemInputStream would report how much data had been buffered, which is not equivalent to the amount of data that is available. Only data that has been followed by and end-of-line sequence or the end-of-stream is available (except in raw mode). Solution : Rewrite SystemInputStream to track the amount of data available to the application. -=============- Problem : java.lang.Double was still unaccurate and slow, but it was also huge. Since : Beta Description: parseDouble() would recognize special values like "inifinity" and "NaN", which the Sun implementation did not do. Values such as 1.23E-300 would print as "1.23E-3", dropping the trailing zeroes. Solution : Rewrote the toString() and parse(String) algorithms to be simpler, faster, and more accurate. -=============- Problem : If someone (like Tom) accidentally typed 'source TiniHttpServer.tini' in at the shell, he would see huge stream of binary data spew with messages that these were all unknown commands. Since : Beta Description: The source command is meant to be run on scripts. If you run it on a TINI file, it doesn't know that the non-ASCII characters mean this file is not a script. Solution : Check the file first for non-ASCII (> 0x07F) characters and abort if it finds any. This could be time consuming if you have a large script file to run, so this check can be disabled with the new '-d' switch to the source command. -=============- Problem : Trying to set the date to anything later than the 25th of the month through TINI's com.dalsemi.system.Clock class (using a long count of milliseconds to set the date) would not work. After the data was written and re-read from the clock, TINI would report the date as the 1st. Since : Beta Description: The algorithm we were using would come up with a negative day of the month if the millisecond being parsed was after the 25th day of the month. When this value was sent to the method that actually wrote values into the clock, it was bounds-checked to 1. Solution : Correct the algorithm to accept any day of the month. -=============- Problem : HTTP POST's through the URLConnection classes did not properly send out any headers before POSTing the data. Since : Beta Description: Before the InputStream is returned to the calling program, the URLConnection should send any headers the user wants sent through the OutputStream to the URL. This was not happening. Solution : Send the headers before giving the outputstream to the calling program. -=============- Problem : TINIOS.serExternalSerialPortEnable method did not enable the port unless it was called for both external ports. Since : Beta Description: This was due to a bug in the native method that enabled the ports. Solution : Fixed the method to enable the requested external port. -=============- Problem : Calling close on an external serial port or remove on an ExternalInterruptEventListener could cause TINI to reboot or other form of death or data destruction. Since : Beta Description: This was a bug in the external interrupt driver which used an uninitialized index to remove a port or listener from the external interrupt ISR chain. Solution : Driver now indexes properly into the ISR chain on removal. -=============- Problem : Internal serial ports consumed memory on repeated open and closes. Since : Beta Description: Driver failed to free a handle on close. Solution : Driver now frees all memory on close. -=============- Problem : java.lang.Math.sqrt returned incorrect results. Since : Beta Description: This was due to a bug in the native sqrt method. Solution : Fixed. -=============- Problem : TINIExternalAdapter onewire adapter would not work correctly with alarming DS1994 parts. Since : Beta 2 Description: DS1994 parts have a special alarm mode that could cause problems communicating with other 1-Wire devices on the network. Solution : Fixed. -=============- Problem : TINIExternalAdapter did not support programming pulse Since : Beta 3 Description: EPROM 1-Wire parts require a 12V programming pulse for write operations. TINI did not support this feature, and therefore could not program EPROM devices. Solution : Added support for programming pulse. -=============- Problem : TINIExternalAdapter and TINIInternalAdapter getSpeed() method would return incorrect results. Since : 1.01 Description: N/A Solution : Fixed. -=============- Problem : TINIInternalAdapter searches can miss some devices. Since : Beta 3 Description: In very rare cases, the internal 1-Wire search could miss some devices. Solution : Fixed. -=============- Problem : com.dalsemi.onewire.util.Address.getAsString() is slow Since : 1.01 Description: N/A Solution : Now three times faster. -=============- Problem : CanBus.receiveFramesAvail() returned incorrect values Since : 1.01 Description: N/A Solution : Fixed. -=============- Problem : I2C module was not correctly performing the stop condition Since : 1.01 Description: N/A Solution : Fixed. -=============- Problem : InetAddress returns localhost hostname incorrectly Since : 1.01 Description: InetAddress would return "localhost" as the hostname in instances when the address was not 127.0.0.1. Solution : Fixed. -=============- Problem : The system clock drifts slow over time. Since : Beta 3 Description: The system clock is the internal time source used for timeouts, thread swaps, process swaps, system uptime, and other similar time related systems. The system clock was drifting slower than wall clock time which caused Thread.sleep() calls to take longer than necessary to wake up, and also gave incorrect uptime figures. Solution : Fixed. -=============- Problem : DataPort FIFO read/write operations to CE space are much slower than sequential. Since : Beta 3 Description: N/A Solution : Fixed, about 3 times faster. -=============- Problem : TINIOS.reboot() would fail to reboot TINI. Since : 1.01 Description: N/A Solution : Fixed. -=============- Problem : TINIConvertor does not allow multiple directories to be specified. Since : 1.01 Description: Using the -f option with multiple directories, or a mixture of directories and files would fail with exceptions. Solution : Fixed, user can specify any number of directories, ZIP files, class files, and jar files on the command line. -=============- Problem : ByteArrayOutputStream is slower than it should be. Since : 1.01 Description: N/A Solution : Fixed. -=============- Problem : CommPortIdentifier.isCurrentlyOwned returns false if port is owned by another process. Since : Beta Description: N/A Solution : Fixed. -=============- Problem : DHCP with WINNT 4.0 DHCP servers. Since : 1.01 Description: WinNT dhcp server would not respond to discover messages from TINI. Packets were one byte too long Solution : Packet lengths were reduced. -=============- Problem : DHCP using different networks. Since : 1.01 Description: TINI would attempt to claim last known ip and due to stale settings would never properly initiate a new discover message and hence without a B18 F0 or a release (ipconfig -r -f) on the previous network, TINI would never get a new dhcp ip. Solution : Bugs fixed in DHCPClient.java. Tested using our network server and win NT server on a separate network. -=============- Problem : SMTP Client would not work properly with multiline responses from the server. Since : 1.01 Description: SMTP Client only expected single line responses, causing failure with some SMTP servers. Solution : com/dalsemi/protocol/mailto/Connection.java was changed to allow multiline responses from an SMTP server. -=============- Problem : Certain large write operations on SocketOutputStream objects cause an IOException. Since : 1.0 Description: N/A Solution : The IOException was generated by a false validation error in the socket layer. This has been fixed for 1.02. -=============- Problem : The ethernet driver doesn't play nicely with Java code wanting to peek at the controller's register contents. Since : Alpha Description: N/A Solution : The ethernet driver has been modified to allow application code to safely read status registers. -=============- Problem : When too many sockets are created, the last underlying TCP connection is not released when the process terminates. Since : 1.0 Description: N/A Solution : The socket layer allowed 1 too many sockets to be created. This is fixed in 1.02. -=============- Problem : The getBoolean method in class java.lang.Boolean is not static. It should be. Since : Beta Description: N/A Solution : getBoolean has been changed to static in 1.02. -=============- Problem : Writing to a closed socket results in either an InterruptedIOException or a SocketException instead of an IOException. Since : Beta Description: N/A Solution : This has been fixed in 1.02. Writing to a closed socket will result in an IOException being thrown. -=============- Problem : The javax.comm.SerialPort DATA_AVAILABLE event fires multiple times for the same data if the data is not read in the event listener. Since : Beta Description: According to the javadocs for this event - This event will be generated once when new data arrive at the serial port. Even if the user doesn't read the data, it won't be generated again until next time new data arrive. Solution : This event now fires only when new data arrives. -=============- Problem : TINI reboots when File.list() is called simultaneously on the same file. Since : Beta 3.0 Description: If the ls command is executed from two different sessions at the same time, TINI will reset itself and some of the files could be lost. Solution : A bug was found and fixed in the firmware. -=============- Problem : Loss of power during file operations can corrupt the file system. Since : N/A Description: Any modifications to the file system could be interrupted by loss of power or by a reset. If the file system was in the middle of critical change, portions of the file system could become corrupt. The file system is "fixed" during the reboot routines, but sometimes at the cost of losing many files. Solution : The file system was changed to handle loss of power. When TINI resets, any critical operations are completed if possible. If not, the file system will be reset to its previous state. -=============- Problem : File.list() for directories with more than 254 files is broken. Since : Beta 3.0 Description: Directories cannot hold more than 254 files. Older versions of TINI firmware allow you to add more than that, but fail when trying to perform a listing. Solution : TINI will not allow more than 254 files in a single directory. An IOException is thrown instead. -=============- Problem : Serial port input buffers were not resizeable Since : Beta Description: Unable to resize the input buffers for the serial ports. Solution : Serial port input buffers can now be resized through javax.comm -=============- Problem : Available method does not return all available bytes on the port. Since : 1.0 Description: Calling the available method on the serial ports does not return all available bytes. Solution : Now returns all available bytes. -=============- Problem : New support for enabling RTS/CTS flow control for the internal serial ports. Since : 1.02 Description: This method must be called to enable RTS/CTS flow control for the internal serial ports. Only one of the internal serial ports may have RTS/CTS flow enabled at a time. If the method is called and RTS/CTS is already enabled on the other internal port, then a javax.comm.UnsupportedCommOperationException will be thrown. Solution : public native static boolean setRTSCTSFlowControlEnable(int portNumber, boolean enable) throws javax.comm.UnsupportedCommOperationException; -=============- Problem : New support for checking if RTS/CTS flow control is enabled on a given internal port. Since : 1.02 Description: This method returns whether RTS/CTS flow control is enabled on a given internal port. Solution : public native static boolean getRTSCTSFlowControlEnable(int portNumber); -=============- Problem : CanBus does not throw exceptions on read/write when bus errors occur. Since : TINI 1.0 Description: N/A Solution : Exceptions will now be thrown for all bus errors -=============-