Re: [pretest] regression in pdftex

On Wed, Feb 18, 2026 at 16:26 (-0700), Max Chernoff wrote:

> Hi Jim,

> On Wed, 2026-02-18 at 10:03 -0400, Jim Diamond wrote:
>> I think a much better choice would be for it to initialize (otherwise
>> uninitialized) automatic variables to, say, -2^14 (or some other value with
>> a good chance of causing an immediately-noticeable problem).

> This is "-ftrivial-auto-var-init=pattern":

> https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html#index-ftrivial-auto-var-init

Max,

I'm glad to see that option exists for the diligent programmer who tests
his/her code thoroughly.

My main concern is about compilers trying to be your friend by having
defaults which cover up errors in your program.  I'd much rather compilers
do what they can to make errors in one's programs immediately blow up in
one's face, so one can get on the job of fixing one's errors.  Because some
day that program might be handed off to some innocent "user" who compiles
it on a different system with a different compiler, and then Bad Things
might happen.  (As we saw recently here.)

A long time ago a second year student in my C programming class had some
uninitialized pointer in his C program.  When I ran it on my system, it seg
faulted and core dumped.  But he was using a different Linux distro than
me, and, interestingly, on his system this uninitialized pointer pointed
into some (apparently) otherwise unused part of his address space, and his
program ran to completion with correct output.  He was rather incensed when
I insisted that his program was nonetheless not correct.

But I suppose this discussion might be more appropriate in a compiler
writer's mailing list.  :-)

Cheers.
                                Jim