On Wed, Dec 24, 2008 at 12:14 PM, Shinya Kuribayashi shinya.kuribayashi@necel.com wrote:
Graeme Russ wrote:
[Snip]
- All weak functions are declared as __function() in the source file with
funtion() __attribute__((weak, alias("function"))); on the line immediately after the closing brace of __function() - for example: void __do_something (args) { ...some code... } do_something(args) __atttribute__((weak, alias("__do_something")));
Why do we need to being consistent? I don't see real/technical benefits to doing so. Using alias or not should be each developer's business. I completely disagree with forcing alias use here.
Consistent code is easier to maintain - If we eliminate inconsistencies early, it stops them being spread and resulting in spaghetti code
[snip]
diff --git a/board/incaip/incaip.c b/board/incaip/incaip.c index 3b30970..3ee3ac9 100644 --- a/board/incaip/incaip.c +++ b/board/incaip/incaip.c @@ -31,7 +31,7 @@
extern uint incaip_get_cpuclk(void);
-void _machine_restart(void) +void machine_restart(void) { *INCA_IP_WDT_RST_REQ = 0x3f; }
Why change the function name? It's derived from Linux/MIPS, and I'd like to keep consistent with with him. Please don't touch it.
For consistency :) - But I 100% agree that if the code is coupled to another source and meets that sources rules then it is easier and more maintainable to keep their rules - I have no objection to retaining the original names
[snip]
diff --git a/cpu/mips/cpu.c b/cpu/mips/cpu.c index b7180b0..84c4730 100644 --- a/cpu/mips/cpu.c +++ b/cpu/mips/cpu.c @@ -38,13 +38,13 @@ : \ : "i" (op), "R" (*(unsigned char *)(addr)))
-void __attribute__((weak)) _machine_restart(void) -{ -} +void inline __machine_restart(void) {} +void inline machine_restart (void)
__attribute__((weak, alias("__machine_restart")));int do_reset(cmd_tbl_t *cmdtp, int flag, int argc, char *argv[]) {
_machine_restart();
machine_restart(); fprintf(stderr, "*** reset failed ***\n"); return 0;Why inline? It seems totally wrong to me.
I wasn't aiming to change the semantics of whether functions where inline or not (only weak / aliased). I agree that inline does not make sense here
[snip]
Notice that in Linux, the 'alias' construction is not being used
massively. Can it be removed here also, or is it somehow mandatory here?
I don't think it is mandatory but it is in the majority in u-boot.
If it's not mandatory, please drop all aliases. That saves source code size (not generated u-boot image size) a little bit. Majority or not does not make sense here.
I wanted to make the impact of the patch as small as possible - I agree that this would make the code cleaner, simpler and smaller but would require more extensive testing across a lot of build-tools to make sure they all fully support weak functions in this way. I am more than happy to create a patch, but it will require a _lot_ of regression testing
Regards,
Graeme