[tlbuild] initial builds for tl'16

[Mojca Miklavec]

On 6 March 2016 at 07:47, Simon wrote:
> Karl,
>
> On Sat, 2016-03-05 at 23:48 +0000, Karl Berry wrote:
>> Hello TeX Live builders.  Today is purportedly the day to begin
>> *testing* of the sources for the TL'16 release.
>>
>> Simon Dales (armhf-linux)(Raspberry Pi)
>
> Main: build + self test: OK
>
> xz + wget: builds OK.
>
> /// asy
>
> Builds OK.
>
> Testing asy complains about by Ghostscript version and stops at that
> point. The tests before that point seen OK. Debians wheezy and jessie
> have 9.05, 9.06 respectively. Asy demands 9.14. This requirement is met
> with stretch and sid, but these aren't available for RPi yet.
>
> What GS features does the test script use that are in 9.14 and not on
> 9.05?

Asy calls GhostScript with
    -sDEVICE=eps2write
which is apparently only available (or reliable?) since 9.14.

>From what I understand GhostScript provided *only* "epswrite" up to
some version and *only* "eps2write" from some version on.

According to http://ghostscript.com/doc/current/Issues.htm epswrite
became deprecated in version 9.12.

When epswrite was gone from GS, asy had to be fixed. The relevant
commit is most likely this one:

https://github.com/vectorgraphics/asymptote/commit/8815ffb186bf0b4237449b4b7a1277abf5751d6d

But there might be other bugfixes (I didn't try to look too hard):

https://github.com/vectorgraphics/asymptote/commit/84499463e395a2953059b659afe614a9f207c083

So asy basically needs to know in advance which version of GhostScript
is being used. For simplicity it simply dropped support for older
versions of GS.

I don't know how hard it is to support both, but it is certainly a
pain, esp. if asy would be compiled on one machine and then executed
on the other with a different version of GS. Then it would have to
keep checking for GS version every time before being able to actually
do any work.

But this has been the case in TL 2015 already.

Mojca