The test gets stuck on Solaris CI.
20211006T050007Z.fail.html.gz#test-all
```
[14558/21042] TestProcess#test_no_curdirtimeout: output interval exceeds 1800.0 seconds.
timeout: the process group 3857 is alive.
```
Related to ee89543e09
After the change to use realpath on loaded features, Solaris CI
started failing in test_no_curdir (which tests behavior for running
ruby without a working directory).
I was able to trace the problem to the following call chain:
rb_call_inits->Init_Thread->Init_thread_sync->rb_provide->
get_loaded_features_index->rb_check_realpath->rb_dir_getwd_ospath->
ruby_getcwd
This will throw an exception, but because Ruby hasn't been fully
initialized at the point the exception is thrown, it just exits
with a status of 1.
The bug here is that rb_check_realpath should not raise an
exception, it should return nil. This bug is hit on Solaris
because Solaris uses the realpath emulation instead of native
realpath, and the realpath emualation raised instead of
returning nil if the mode was RB_REALPATH_CHECK. Use rb_rescue
in the realpath emulation if the mode is RB_REALPATH_CHECK, and
swallow any exceptions raised and return nil.
This fixes issues with paths being loaded twice in certain cases
when symlinks are used.
It took me multiple attempts to get this working. My original
attempt tried to convert paths to realpaths before adding them
to $LOADED_FEATURES. Unfortunately, this doesn't work well
with the loaded feature index, which is based off load paths
and not realpaths. While I was able to get require working, I'm
fairly sure the loaded feature index was not being used as
expected, which would have significant performance implications.
Additionally, I was never able to get that approach working with
autoload when autoloading a non-realpath file. It also broke
some specs.
This takes a more conservative approach. Directly before loading the
file, if the file with the same realpath has been required, the
loading of the file is skipped. The realpaths are stored as
fstrings in a hidden hash.
When rebuilding the loaded feature index, the hash of realpaths
is also rebuilt. I'm guessing this makes rebuilding process
slower, but I don think that is a hot path. In general, modifying
loaded features is only done when reloading, and that tends to be
in non-production environments.
Change test_require_with_loaded_features_pop test to use 30 threads
and 300 iterations, instead of 4 threads and 1000 iterations.
I saw only sporadic failures with 4/1000, but consistent failures
30/300 threads. These failures were due to the fact that the
concurrent deletions from $LOADED_FEATURES in other threads can
result in rb_ary_entry returning nil when rebuilding the loaded
features index.
To avoid concurrency issues when rebuilding the loaded features
index, the building of the index itself is left alone, and
afterwards, a separate loop is done on a copy of the loaded feature
snapshot in order to rebuild the realpaths hash.
Fixes [Bug #17885]
In certain conditions, Regexp#match could return a MatchData with
missing captures. This seems to require at the least, multiple
threads calling a method that calls the same block/proc/lambda
which calls Regexp#match.
The race condition happens because the MatchData is passed from
indirectly via the backref, and other threads can modify the
backref.
Fix the issue by:
1. Not reusing the existing MatchData from the backref, and always
allocating a new MatchData.
2. Passing the MatchData directly to the caller using a VALUE*,
instead of indirectly through the backref.
It's likely that variants of this issue exist for other Regexp
methods. Anywhere that MatchData is passed implicitly through
the backref is probably vulnerable to this issue.
Fixes [Bug #17507]
Pass in ec to vm_opt_newarray_{max,min}. Avoids having to
call GET_EC inside the functions, for better performance.
While here, add a test for Array#min/max being redefined to
test_optimization.rb.
Fixes [Bug #18180]
The vi mode can handle "argument number" before an operator or a motion,
such as, "3x" (equals "xxx"), and "3l" (equals "lll"). In the emacs
mode, GNU Readline can handle argument number with meta key, like
"Meta+3 x" (equals "xxx").
9183cc2e8b
%v is supposed to be the VMS date, and VMS date format uses an
uppercase month.
Ruby 1.8 used an uppercase month for %v, but the behavior was
changed without explanation in r31672.
Time#strftime still uses an uppercase month for %v, so this change
makes Date#strftime consistent with Time#strftime.
Fixes [Bug #13810]
56c489fd7e
Do not change the password of the input parameter source during
anonymization, by using the proper helper instead of changing the
original uri directly.
eaa2dd8a97
Implement deep cloning for `Gem::Uri` class to fix a bug where redacting
credentials modifies the URI string in place instead of returning a
modified copy.
eafb5a279b
