Re: [Tinycc-devel] asm labels

[Thomas Preud'homme]

Le lundi 13 septembre 2010 18:04:11, grischka a écrit :
> Thomas Preud'homme wrote:
Oh no, I forgot again to reply to the list also. Sorry.
> > I don't know why it has been invented for (you could take a look at git
> > log on glibc to find the commit which introduced this) but in (e)glibc
> > the way they use it bypass the type checker. That is they declare
> > functions cosl to be cos and there is no warning if cos is used with an
> > argument long double instead of double. As preprocessing is done before
> > type checking there is a difference. As to why it is called asm label
> > it's that it's suppose to change the name used in the generated asm.
> > Macro substitute at source level whereas asm label change the name at
> > asm level. I agree it's not what I did in the patch but it was simpler
> > (and thus much smaller) that way. I don't know how gcc guys did it.
> 
> Sigh, what can we do when gcc decided to have it ;)
It's more about people assuming gcc feature are presents on all compiler 
although it doesn't make a difference in the facts.
> 
> >> Anyway, as to tinycc, I noticed that it fails to compile with
> >> NOTALLINONE defined.  So if this is not an "asm-feature" really I'd
> >> suggest to move the two functions "parse_asm_str" and "asm_label_instr"
> >> elsewhere (e.g. to libtcc.c or tccgen.c) and remove CONFIG_TCC_ASM from
> >> tccasm.c again. The idea is that tccasm.c can stay out from compilation
> >> on platforms without asm support.
> > 
> > Do you have a regression test to detect this kind of regression or you
> > just felt this patch would break compilation with NOTALLIONE ? If you
> > have some kind of automated test I'm very interested to use it to
> > produce better patches.
> 
> Well, I'm using NOTALLIONE personally (of course, since I've put some
> effort to make it work).  So I saw it broke and I felt that it could
> break on platforms without asm support (e.g. arm) even without NOTALLIONE
> because "tccasm.c" wouldn't be included at all.
I tried on arm and it worked.
> 
> Maybe we should make "NOTALLIONE" (= compilation from single objects) the
> default on the native platform now.  ("Makefile" does not yet support it
> it for cross compilers).
> 
> > I don't think it does matter whether it is an asm feature or not. The
> > fact is it's independant from the asm language so yes it could be moved.
> > And more other, I don't think CONFIG_TCC_ASM_LABEL is really useful
> > after all as this code is platform independant. I'll move
> > asm_label_instr and parse_asm_str out of tccasm.c to put tccasm.c as it
> > was previously. By the way, should I rename the two functions above to
> > avoid confusion ?
> 
> Name new function as you see fit, rename existing function only for
> strong reasons ;)
That's a good advice. Thanks
> 
> --- grischka
Thomas Preud'homme

signature.asc
Description: This is a digitally signed message part.