Re: final steps for reproduceable builds.

["Perry E. Metzger" <perry%piermont.com@localhost>]

christos%astron.com@localhost (Christos Zoulas) writes:
> In article <87ljredwx1.fsf%snark.cb.piermont.com@localhost>,
> Perry E. Metzger <perry%piermont.com@localhost> wrote:
>>
>>"Perry E. Metzger" <perry%piermont.com@localhost> writes:
>>> I'll conditionalize the ar stuff on AR_DFLAG being set to yes and set
>>> that in build.sh in the manner of BUILDSEED, how's that?
>>
>>New patch (untested -- I'm running a build now) -- comments?
>>
>>Perry
>
> 1. this makes the assumption that our ar does ranlib, which is only true
>    for gnu ar and will break other toolchains.

It appears that POSIX now more or less requires that ar do ranlib, and
requires that there be a -s flag that does what gnu ar's s flag
does.

Our old "ar" was not POSIX, but if we ever re-write it to be modern,
it will need -s anyway, with the stated effects.

See:

   http://www.opengroup.org/onlinepubs/009695399/utilities/ar.html

It is true that we may break very old toolchains that are not POSIX
but I don't see that as a problem. (We also would break toolchains
that don't supply "D" but one can always set AR_DFLAG to "no" -- the
other options are all POSIX.)

> 2. the yes or no variable names have MK in front of them typically, the
>    AR_DFLAG one makes me think I should set it to D or nothing.

I know, but it doesn't really seem to fit into the naming scheme very
organically. We're not making or not making something based on the
flag. The name bothers me too -- both ways.

Perry
-- 
Perry E. Metzger                perry%piermont.com@localhost