rb_fstring unsafe to use in general case
I don't think the current "frozen"f in pure Ruby code is unsafe
(I haven't reproduced a crash)
But I do not think rb_fstring is generally safe for other C code, including extension authors
Consider the following case:
fstr1 = rb_fstring(str) fstr1 goes out of scope GC mark runs ... fstr1 is eligible for lazy sweep fstr2 = rb_fstring(str) fstr2 is identical to fstr1 fstr1 is swept (rb_str_free) fstr2 use attempted -> crash
I extracted this bug report from Feature #8998, where I showed a patch
which exposes this bug during "make check"
#2 [ruby-core:57800] Updated by normalperson (Eric Wong) about 4 years ago
"ko1 (Koichi Sasada)" email@example.com wrote:
Nobu's patch r43210 may solve this problem. Could you verify it?
Thanks, I did not think to update my repo :x
Anyways, r43210 works fine.
BTW, I don't like this fix. I'll try fix by another way.
I'm not sure why you don't like it, but I'll be glad to test your fix.