pass_comments misinterpreted URL in a string-literal as a label+ comment

[erichkeane]

So I found this problem trying to minimize something that included libstdc++.

The original code was:

static_assert(sizeof (uintmax_t) == sizeof (unsigned long long),
"This library calls __builtin_clzll on uintmax_t, which "
"is unsafe on your platform. Please complain to "
"http://gcc.gnu.org/bugzilla/");

the pass replaced the last line there with:
"http:

(note no closing string either!).

The result is it spent all weekend with no ability to clean up anything.

[regehr]

Hi Erich, I don't understand why this made it take all weekend, can you explain more?

But anyway, the model here is that C-Reduce can try anything at all, and your interestingness test should be robust with respect to this.

[erichkeane]

I don't particularly know how creduce works, but I would presume because it resulted in no other passes being able to make any progress, since no change it made could actually pass the script.

I let it run over the weekend, and it removed no other code, and I had to kill it. It never ended. "top" showed the clang instances coming and going, but no additional output happened. I saw this:

===< pass_unifdef :: 0 >===
===< pass_comments :: 0 >===
(0.0 %, 7376320 bytes)
===< pass_ifs :: 0 >===
===< pass_includes :: 0 >===
===< pass_line_markers :: 0 >===
===< pass_blank :: 0 >===
===< pass_clang_binsrch :: replace-function-def-with-decl >===
===< pass_clang_binsrch :: remove-unused-function >===
===< pass_lines :: 0 >===

[erichkeane]

As determined in the second bug report (#217), the 'hanging' behavior that made it take all weekend was a result of my script doing the file name with the full path (since I mis-interpreted the initial interestingness test error message).

So, this sentence: "The result is it spent all weekend with no ability to clean up anything." in my initial bug report was a different issue.