[tlbuild] dvi test?

[John Hawkinson]

Good evening/morning. I haven't forgotten about trying to narrow down the root cause for why MacOS is different, although with a solid workaround in place it has definitely not been a priority. (I did try some things unsuccessfully.)

Paul Vojta <vojta at math.berkeley.edu> wrote on Fri, 11 Feb 2022
at 23:17:16 EST in <Ygc0zLtxA7KJzjU8 at math.berkeley.edu>:

> Could someone try this out (on a Mac) and check whether the seek operation
> still fails?  (This is relative to the latest dvi-init.c in the TeX Live
> repository, from about 10 days ago.)

Yes, it does! That is, without the patch, r61850 produces:

jhawk at lrr xdvik % TEXMFROOT=/usr/local/texlive/2021 TEXMFCNF=$TEXMFROOT/texmf-dist/web2c  ./xdvi-bin t2a
xdvi-bin: Error: Couldn't seek to start of file /var/folders/m2/16zlwjgj2p5d3qdy99z2gzf40000gn/T/xdvi-oAjS3R: Invalid argument - disabling `useTempFp'; target_fp: 0x1dfbeed40.

And adding your patch to use dup() then produces no such output.

> Also, I noticed that reading the pages from the dvi file is done entirely,
> or at least mostly, with kernel calls (read() and lseek()), so I guess
> it's not surprising that the libc state is corrupted.

Huh! This is...xxone() I guess?

It does still leave open why it works reliably on nearly all OSes but not recent MacOS, but that does tend to suggest this not Apple's fault. Might make it harder to get the "bug" fixed in the OS (I do wonder how much other software might break in this same way...probably not a whole lot...)

--
jhawk at alum.mit.edu
John Hawkinson