[Bug middle-end/78901] [7 Regression] ICE: verify_gimple failed (error: statement marked for throw in middle of block)

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78901

--- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 40404
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=40404&action=edit
gcc7-pr78901.patch

There are multiple issues, this patch hopefully fixes just one of them - the
case when the call might throw and is replaced by a constant that can't throw.
The other problems are:
if the call is not the NULL buffer, 0 size case, where the call is kept, just
lhs removed and lhs = cst; added after the call, then if the call might throw
(thus has to be last in the bb), inserting a stmt after it is invalid - if
(stmt_ends_bb_p (info.callstmt), then the stmt needs to be added to the
beginning of the fallthru basic block.  And another problem is that while you
nicely remove lhs = snprintf (0, 0, ""); call and just set lhs = 0;, if the
user just writes useless snprintf (0, 0, ""); without lhs, then nothing removes
it at all.  If the call has no lhs and is info.nowrite, then I think we can
just remove it (the case replacing it with constant will work for that too).