This is the mail archive of the mailing list for the binutils project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] GNU_AR_DETERMINISTIC env var, --enable-deterministic-archives

Hi Roland,

I agree with this as a general principle.  (But it's worth pointing out
that GCC is indeed affected by various environment variables--which, at
least, are documented in their own section of the GCC manual.)


Well, it's just not very true at all about GCC.  But I take your point.
However, you didn't say anything about the build-time configure option.

Sorry, I should have covered that.

I am OK with the configure option. In fact I think that it is the best way forward. Allow users (or toolchain providers) to choose what the default behaviour will be, and provide a command line option to explicitly determine determinism.

Also - there needs to be some way for the user to find out how the
particular ar/ranlib that they are using was configured.

What do you have in mind? It's easy enough to add the information to --help output,

That was what I had in mind.

but that is not programmatically friendly.

True. Another option would be to have "ar --verbose" display the configure arguments that were used when it was built. Ie similar to the way that gcc behaves.

But I don't
really see why someone really needs to know the default.

I was thinking more about debugging again. If user reports a strange behaviour with ar not storing file timestamps say, then we can ask them to run "ar --help" or "ar --verbose" and explain what the output means.

So - if you will accept dropping the environment variable and just using a configure time parameter to set the default behaviour then please do resubmit your patch.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]