My previous commit, c6d2c81, broke Keil builds and maybe even IAR.
I need to learn how to read C code :) I thought I was masking off the
O_BINARY bit only for GCC builds but it turns out that my update was
in the fall-through case for all toolchains. This commit now places
the O_BINARY masking operation into a GCC specific #ifdef clause.
Testing:
I tested the same fopen("/local/out.txt","rb") code as before but this
time I built it with the online compiler and GCC_ARM. I tested the
resulting binaries on mbed-LPC11U24 and mbed-LPC1768 boards.
Thanks to @neilt6 for catching & reporting this!
Fixes issue #1562 reported by @justbuchanan.
When building code with GCC-ARM / newlib, attempting to use the
b (binary) mode in a fopen() call would fail. newlib would parse
this option and pass it down to the LocalFileSystem open call which
didn't like the unexpected O_BINARY flag in openmode.
The openmode_to_posix() function in retarget.cpp would never set the
O_BINARY flag for the other toolchains but for GCC it would just pass
down whatever newlib placed there. This commit masks out the O_BINARY
bit so that it never gets passed down to the file system drivers, just
as occurs for the other supported toolchains.
Test case:
FILE *fp = fopen("/local/out.txt", "rb");
I tested that code on mbed LPC1768 and LPC11U24 boards while using
GCC_ARM as the toolchain. It failed on both platforms previous to
this change and succeeded there after.
I moved the exit handler from exit.c to retarget.cpp where the rest of
the standard C library retargetting routines are found. The exported
makefiles already explicitly link in retarget.o but not exit.o
When building with the GNU toolchain, it is safest to link in the
object file directly for retargetting routines so that GNU linker has
to use these versions of the routines.
Testing:
I have only tested this change with GCC_ARM. It should really be
tested with more toolchains before getting merged.
A recent commit, 43acaa4166, to get _sbrk() to build successfully for
LPC2368 broke the Cortex-M implementation. __get_MSP() isn't ever
defined as a macro, it is an inline function. This means that the
code would always be compiled to use SP instead of MSP on Cortex-M
parts. I switched the code to instead use the TARGET_ARM7 define
to choose which stack pointer to utilize.
I tested this fix by making sure that the LPC2368 version of the mbed
SDK would still build successfully with the Python scripts and that the
NET1 test still built and ran successfully on my mbed-LPC1768 device.
If the FileBase::lookup operation in the constructor of FilePath returns
NULL, subsequent operations (such as isFile()/isFileSystem()) will call
methods on a NULL 'fb' pointer. This commit fixes this issue by adding
explicit NULL checks and a new method in FilePath (exists()).
I verified that the hang issue I was seeing when building and running
the mbed official networking tests with GCC_ARM was related to this
issue reported on the mbed forums:
http://mbed.org/forum/mbed/topic/3803/?page=1#comment-18934
If you are using the 4.7 2013q1 update of GCC_ARM or newer then it
will have a _sbrk() implementation which checks the new top of heap
pointer against the current thread SP, stack pointer.
See this GCC_ARM related thread for more information:
https://answers.launchpad.net/gcc-arm-embedded/+question/218972
When using RTX RTOS threads, the thread's stack pointer can easily
point to an address which is below the current top of heap so this
check will incorrectly fail the allocation.
I have added a _sbrk() implementation to the mbed SDK which checks the
heap pointer against the MSP instead of the current thread SP. I have
only enabled this for TOOLCHAIN_GCC_ARM as this is the only GCC based
toolchain that I am sure requires this.
- rename stdio.cpp to retarget.cpp, since it doesn't contain only stdio-related
code
- move __cxa_pure_virtual from exit.c to retarget.cpp where it belongs.
Adam Green
sg-
Masao Hamanaka
0xc0170
GustavWi
Martin Kojtal
Alexander Valitov
Andreas Løve Selvik
Bogdan Marinescu
dinau
Bogdan Marinescu
Emilio Monti