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

Re: Contributing to pandoc dependencies: many packages to upload



On Tue, 31 May 2022, Robert Greener wrote:

In thinking about the packages that are FTBFS when using
LC_ALL=C, this is
cause by somewhat of a regression in haskell-devscripts, since
the old
install recipe used to set LC_ALL=C.UTF-8[1] before running the
command
where most of these packages are failing (due to the copyright
symbol).

I'm thinking that rather than update all of these individual
packages (and
who knows how many more) to set LC_ALL=C.UTF-8, we should restore
this
setting in haskell-devscripts, and perhaps even set LC_ALL=C.UTF-
8
globally in hlibrary.mk?

What do you all think?

I think it should be set as minimally as possible - and clearly
documented.

E.g. only within the rules known to specifically need it, not
loaded
into the global make environment across all make targets.

That was my initial thinking, too.  However, the buildd's set
LC_ALL=C.UTF-8, so wouldn't we want packages built outside the
buildd's to
be built consistently?

I think it might make sense to be as minimal as possible. I think the
reproducibility tests alternative locales (e.g., fr_CH.UTF-8). If we
force this for everything it would make it impossible to run buildtime
tests in different locales.

If it is fixed minimally, the haskell-filestore package will need it
setting in debian/rules as the tests explicitly test non-ascii
filepaths.

For reference, there are 164 packages that FTBFS that are maintained by
the haskell team.

Yikes.  Where did you get that number?

Scott

Reply to: