Re: enforcing an UTF8 locale while building a package
- Date: Wed, 10 Jan 2018 23:41:13 +0000
- From: Simon McVittie <smcv@xxxxxxxxxx>
- Subject: Re: enforcing an UTF8 locale while building a package
On Tue, 09 Jan 2018 at 19:52:23 +0530, Chris Lamb wrote:
> Just as one example, a timezone library that did not work properly
> in timezones beyond UTC+0800, etc.
That's a bug, sure, but is it necessarily a release-critical bug? The
answer is usually "it's up to the maintainer", whereas FTBFS on a
reasonable system under reasonable conditions is automatically RC by
policy, and even if it wasn't, FTBFS on a buildd whose architecture
previously built OK has RC-like implications on testing migration.
If we held, for instance, gcc to this standard ("gcc must work properly
with all valid C inputs") then it would have a lot more RC bugs. Instead,
we accept that, while a bug-free compiler would be great, we aren't
going to have one any time soon, and expect the maintainers of other
software to work around bugs with -O0 or code changes if they need to.
A timezone library should be running its unit tests in lots of timezones
whether it's on the reproducible builds infrastructure or not, but
a tangentially related piece of software that happens to use a timezone
library during build shouldn't be responsible for verifying that its
timezone library works in all reasonable situations.
(Note the weasel word "reasonable", of course - arguably our buildds are
already unreasonable in various ways that make sense for compilation,
but less so for testing.)