[Reproducible-builds] GCC patch reviewed. Proposed mail for gcc-patches mailing list

[Santiago Vila]

On Mon, Nov 09, 2015 at 01:02:40AM +0100, Dhole wrote:
> It would be very beneficial to our project (and other free software
> projects working on reproducible builds) if gcc supported this feature.

Hi.

I have very mixed feelings about this kind of patches.

I fear that by modifying gcc to hide the improper usage of __DATE__
and __TIME__, we could be removing an incentive for maintainers and
authors to write software which is truly reproducible, i.e. we run
the risk of people thinking in the line of "Oh, I will not care about
my program using __TIME__ or __DATE__ because gcc will take care of
that".

I also think that this patch will not really make any software to be
reproducible, and it will only serve to "make up" the statistics.

If you build a package twice with llvm and the results differ, then I
would consider the package to be still unreproducible, no matter how
much we patch gcc.

Sure, we can patch llvm as well, but really, what is at fault here is
the program using __DATE__ and __TIME__, not the compiler.

So, I don't think that this patch would really be "beneficial to our
project", as it will only serve to artificially "improve" the statistics.

It would be quite paradoxical indeed that we "make Debian reproducible"
while the software that we distribute is still not reproducible by itself.

"So, do you propose to modify and patch those 500 packages instead?" you might ask.

Well, I do not propose anything as we are all volunteers and nobody
can tell anybody else what he/she should do, but if I were alone in
this project, that would be my goal, yes, even if it takes a lot more
time than patching a single package (gcc). To me, that would be the
right thing to do.

Thanks.