[tlbuild] upmendex, U_INVALID_FORMAT_ERROR, U_ICUDATA_ENTRY_POINT, isBigEndian

[Nikola Lečić]

On Mon, 4 Apr 2016 23:26:23 +0000
  in <201604042326.u34NQN6O000393 at freefriends.org>
  Karl Berry <karl at freefriends.org> wrote:

> Regarding the failure of upmendex.test on 32-bit freebsd et al.
> 
> The short answer: the workaround I found is to explicitly specify the
> compiler (both CFLAGS and CXXFLAGS) flag -DU_IS_BIG_ENDIAN=0.
> (This should not be needed in any other environment.)

It was noticed that compiling with these flags change all binaries:
http://tug.org/pipermail/tlbuild/2016q2/003495.html

I've just noticed that (at least latex and xetex) binaries compiled with
these flags are significantly (almost ~45%) slower than those compiled
without them.

Three xelatex runs on texlive-sr.tex:
with -DU_IS_BIG_ENDIAN=0:
  109.005u 0.842s 1:39.62 110.2% 26398+4722k 0+13io 0pf+0w
without:
  64.955u 0.897s 1:01.01 107.9%	26147+4603k 0+17io 0pf+0w

Four latex+tex4ht runs on texlive-sr.tex:
with -DU_IS_BIG_ENDIAN=0:
  14.835u 0.323s 0:15.17 99.8% 2526+1378k 3+44io 36pf+0w
without:
  10.381u 0.409s 0:10.79 99.9% 1856+1229k 0+44io 0pf+0w

This is a pretty significant difference. Do you have any idea why this
happens? Can other builders that used -DU_IS_BIG_ENDIAN=0 comfirm this?

Anyway, it seems I'll have to rebuild all FreeBSD binaries, except
upmendex. We don't need TL2016 which is almost 50% slower than TL2015.

-- 
Nikola Lečić = Никола Лечић
fingerprint : FEF3 66AF C90E EDC3 D878  7CDC 956D F4AB A377 1C9B
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 314 bytes
Desc: not available
URL: <http://tug.org/pipermail/tlbuild/attachments/20160501/4374e0ea/attachment.bin>