Op 14-10-2025 om 21:52 schreef Eric Raible:
TCCStates should (obviously) be completely independent.
But this corner case shows that a syntax error in the first
one prevents subsequent TCCStates from failing on *correct code*.
My guess would be that a longjmp while processing the syntax
error in the global declaration skips the reset of something associated
with symbol tables.
This is as far as I got: In the non-error case, the first name that
relocate_syms() finds is "memset". In the error case, the first name
that it finds is "L.2". That's probably a good clue for people smarter
than me. The memset is probably required to initialize the
stack-allocated
struct, but for some reason things work when that struct is explicitly
declared (vs. initialized inline).
The code below shows the error when run with an argument.
It works as expected with no arguments. [I apologize for any
formatting errors, I suck at getting gmail to do what I intend.]
I pushed a patch to mob.
The global variable 'global_expr' in tccgen.c was not initialized every
time.
Herman
_______________________________________________
Tinycc-devel mailing list
[email protected]
https://lists.nongnu.org/mailman/listinfo/tinycc-devel
