[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Tinycc-devel] can't build mob branch on x86 64

From: Joe Soroka
Subject: Re: [Tinycc-devel] can't build mob branch on x86 64
Date: Fri, 4 Feb 2011 17:21:42 -0800

On Fri, Feb 4, 2011 at 3:49 PM, Thomas Preud'homme <address@hidden> wrote:
> Really? You want to rewrite it entirely? Took me so much time and it work well
> now :( Couldn't just some CONFIG_TCC_ALLOCA macro be defined in Makefile when
> compiling?

Rewrite it entirely, no.  I apologize for accidentally implying that.
While there *are* problems with your patch (doesn't handle size==0,
doesn't work with bcheck, some others) I don't mean to throw it out.

I guess I just mean that the fact you missed the "TOK_alloca is an
enum" bit plus a few other simple things makes me worry about how well
tested the *complicated* stuff is and I'd kinda like to see you pull
it back, sleep on it, think on it, polish it a little bit and then
bring it back.  I went through the same kind of thing with my patch
for weak symbols - I had to throw it out and start over and ended up
with a much smaller, much better version.

Plus, I had my own partially-working VLA patch in my queue that "took
me so much time" and I hate to throw it out as well, so I know what it
feels like. :)

> Anyway, however you code VLA, you need alloca and need a way to detect if
> alloca is available or not. Do you have suggestion on how to rewrite it?

I have a few ideas, but basically if libtcc1.a doesn't have an alloca
for the target, (or perhaps CONFIG_USE_LIBGCC) then don't do it.
That's the safest way I can think of, at least for tccgen.c.  For
tcctest, you'd need an ifdef that works with gcc too, or modify the
test Makefile.  I'll try to come up with a patch now.

Also, it wouldn't hurt to introduce a __builtin_alloca and use that
for stuff like VLA, so there's no link-time confusion with any other
libc allocas, if they exist.  That gawk report about missing
"__builtin_malloc" after the introduction of VLA seems to hint at that
as a problem.

I dunno, the whole build/test/cross system is a bit of a mess.  I'm
leaning towards my own friendly-fork a-la landley which focuses solely
on x86 and gets it working perfectly there, without regard for other
targets, because I have a lot more patches and with each one I'm
afraid of stomping on something I can't test.

reply via email to

[Prev in Thread] Current Thread [Next in Thread]