Hi,
On 2024-04-09 15:33:10 -0700, Andres Freund wrote:
> Which leads to a control flow graph with *many* incoming edges to the basic
> block containing the function call to mm_strdup(), triggering a normally
> harmless O(N^2) or such.
With clang-16 -O2 there is a basic block with 3904 incoming basic blocks. With
the hacked up preproc.y it's 2968. A 30% increase leading to a doubling of
runtime imo seems consistent with my theory of there being some ~quadratic
behaviour.
I suspect that this is also what's causing gram.c compilation to be fairly
slow, with both clang and gcc. There aren't as many pstrdup()s in gram.y as
the are mm_strdup() in preproc.y, but there still are many.
ISTM that there are many pstrdup()s that really make very little sense. I
think it's largely because there are many rules declared %type <str>, which
prevents us from returning a string constant without a warning.
There may be (?) some rules that modify strings returned by subsidiary rules,
but if so, it can't be many.
Greetings,
Andres