Boris Kolpackov:
> Ximin Luo <infini...@pwned.gg> writes:
>
>> -I to an absolute path is not that common for system / distro-built
>> stuff.
>
> Ok, thanks for clarifying.
>
>
>> In the cases that it occurs, indeed it could and should be fixed
>> by the package buildsystem, e.g. by stripping a prefix when they
>> add -I flags to CFLAGS. But that's a separate issue from what
>> we're talking about here.
>
> I believe it is the same issue: any package that blindly embeds
> information about how it was built into the result of the build
> does not care about reproducible builds.
>
I don't think that's realistic position to take, and it seems like a fairly
arbitrary assertion made merely to "brush this issue under the carpet". I think
this because of the "implicit contract" of how CFLAGS is generally used:
CFLAGS (and other flags) is a way to pass information from multiple layers of
programs down to GCC. It is necessary to "blindly embed" it if you want to save
it *at all*, otherwise you are forced to violate the separation-of-concerns
between layers in your buildsystem stack.
- Program A sets CFLAGS += --flags_A, then calls program B
- Program B sets CFLAGS += --flags_B, then calls program C
- Program C sets CFLAGS += --flags_C, then calls Make
- Make reads CFLAGS then calls $(CC) $(CFLAGS) [etc]
In general, you are not supposed to edit what was already in CFLAGS, you are
supposed to append to it. That is why it's good practise to add "--no" options
for various command-line flags in gcc (and other tools).
So each program X has its own flags_X that it is "responsible for". If all
programs in the stack makes sure their own flags_X contain reproducible values,
then the whole build will be reproducible even when recording CFLAGS.
However, this breaks in the case of -f*-prefix-map. If Program A sets this flag
in flags_A, then this interferes with Program C, even though flags_C was fully
reproducible.
Note that program B and C don't even know that they are being used with GCC,
and so really have no business filtering -f*-prefix-map out of the CFLAGS that
program A already set.
My proposal would allow all the flags_{A,B,C} to be reproducible, and avoid
different programs having to embed layer-violating logic into themselves.
> If others do agree that this should be "fixed" in GCC, then I
> would suggest that you add a separate option for the environment
> variable case (e.g., -ffile-prefix-map-env) and sidestep the
> whole "how to portably distinguish a path from an environment
> variable" issue.
>
That would also be fine for me sure.
X
--
GPG: ed25519/56034877E1F87C35
GPG: rsa4096/1318EFAC5FBBDBCE
https://github.com/infinity0/pubkeys.git
