[tlbuild] dvi test?

[Bruno Voisin]

> On 31 Jan 2022, at 21:35, Richard Koch <koch at uoregon.edu> wrote:
> 
> I just tried
> 
> 	xdvi -notempfile koch.dvi
> 
> with this file. Now R works and the file does not crash. Does this solve the problem on other dvi files?

Yes it does (on koch.dvi and t2a.dvi).

I have been spending way too much time today trying to think of an explanation. But it all comes down to this eventually: why would things be different in recent macOS vs older macOS, when running exactly the binary software on exactly the same binary file? 

The error I noticed earlier in the dtruss output (#22 for lseek) is mentioned on the web as often associated with initialization problems. Which would be consistent with Jonathan Kew's remark last week that "it sounds [...] like there's likely an inadvertent use of an uninitialized (or already-freed) value somewhere, such that the behavior ends up depending on the random contents of some bit of memory."

But there are no such problems on other OSes, or older macOS, for all the years of existence of xdvi. Maybe a change in number types between old and new macOS, yielding different results when performing the arithmetics that generates the arguments of the various lseek calls in dvi-draw.c?

Or this may be totally unrelated, and error #22 of no consequence whatsoever.

Bruno