[#107765] [Ruby master Bug#18605] Fails to run on (newer) 32bit Windows with ucrt — "lazka (Christoph Reiter)" <noreply@...>

Issue #18605 has been reported by lazka (Christoph Reiter).

8 messages 2022/03/03

[#107769] [Ruby master Misc#18609] keyword decomposition in enumerable (question/guidance) — "Ethan (Ethan -)" <noreply@...>

Issue #18609 has been reported by Ethan (Ethan -).

10 messages 2022/03/04

[#107784] [Ruby master Feature#18611] Promote best practice for combining multiple values into a hash code — "chrisseaton (Chris Seaton)" <noreply@...>

Issue #18611 has been reported by chrisseaton (Chris Seaton).

12 messages 2022/03/07

[#107791] [Ruby master Bug#18614] Error (busy loop) inTestGemCommandsSetupCommand#test_destdir_flag_does_not_try_to_write_to_the_default_gem_home — duerst <noreply@...>

Issue #18614 has been reported by duerst (Martin D端rst).

7 messages 2022/03/08

[#107794] [Ruby master Feature#18615] Use -Werror=implicit-function-declaration by deault for building C extensions — "Eregon (Benoit Daloze)" <noreply@...>

Issue #18615 has been reported by Eregon (Benoit Daloze).

11 messages 2022/03/08

[#107832] [Ruby master Bug#18622] const_get still looks in Object, while lexical constant lookup no longer does — "Eregon (Benoit Daloze)" <noreply@...>

Issue #18622 has been reported by Eregon (Benoit Daloze).

16 messages 2022/03/10

[#107847] [Ruby master Bug#18625] ruby2_keywords does not unmark the hash if the receiving method has a *rest parameter — "Eregon (Benoit Daloze)" <noreply@...>

Issue #18625 has been reported by Eregon (Benoit Daloze).

13 messages 2022/03/11

[#107886] [Ruby master Feature#18630] Introduce general `IO#timeout` and `IO#timeout=`for all (non-)blocking operations. — "ioquatix (Samuel Williams)" <noreply@...>

Issue #18630 has been reported by ioquatix (Samuel Williams).

28 messages 2022/03/14

[#108026] [Ruby master Feature#18654] Enhancements to prettyprint — "kddeisz (Kevin Newton)" <noreply@...>

Issue #18654 has been reported by kddeisz (Kevin Newton).

9 messages 2022/03/22

[#108039] [Ruby master Feature#18655] Merge `IO#wait_readable` and `IO#wait_writable` into core — "byroot (Jean Boussier)" <noreply@...>

Issue #18655 has been reported by byroot (Jean Boussier).

10 messages 2022/03/23

[#108056] [Ruby master Bug#18658] Need openssl 3 support for Ubuntu 22.04 (Ruby 2.7.x and 3.0.x) — "schneems (Richard Schneeman)" <noreply@...>

Issue #18658 has been reported by schneems (Richard Schneeman).

19 messages 2022/03/24

[#108075] [Ruby master Bug#18663] Autoload doesn't work with fiber context switch. — "ioquatix (Samuel Williams)" <noreply@...>

Issue #18663 has been reported by ioquatix (Samuel Williams).

10 messages 2022/03/25

[#108117] [Ruby master Feature#18668] Merge `io-nonblock` gems into core — "Eregon (Benoit Daloze)" <noreply@...>

Issue #18668 has been reported by Eregon (Benoit Daloze).

22 messages 2022/03/30

[ruby-core:107819] [Ruby master Feature#18615] Use -Werror=implicit-function-declaration by default for building C extensions

From: "Eregon (Benoit Daloze)" <noreply@...>
Date: 2022-03-10 11:37:25 UTC
List: ruby-core #107819
Issue #18615 has been updated by Eregon (Benoit Daloze).


shyouhei (Shyouhei Urabe) wrote in #note-8:
> There is nothing that a Ruby _user_ can do for this situation.  Gem _authors_ should be aware of their bugs.  Hiding compiler warnings from their eyes is the problem.  `bundle install` is irrelevant here.  I'm not against bundler to remain silent.

Good point, and I think it also makes sense `gem install` does the same (but anyway probably very few use `gem install` directly).
For gem authors, `rake compile` (or manually `cd ext/... && ruby extconf.rb && make`) does show C warnings, so that's fine.

However, gem authors can't predict which functions will be removed in future versions of Ruby (e.g., if the gem is released 2 years before, or if deprecated warnings are hidden by default ...).

IMHO it makes sense to remove deprecated functions after they've been emitting deprecation warnings.

My conclusion is the same: `-Werror=implicit-function-declaration` makes for a better and clearer error for everyone, and there seems no value to let the C extension compile but fail at runtime.

----------------------------------------
Feature #18615: Use -Werror=implicit-function-declaration by default for building C extensions
https://bugs.ruby-lang.org/issues/18615#change-96748

* Author: Eregon (Benoit Daloze)
* Status: Open
* Priority: Normal
----------------------------------------
Currently, if a C extension refers a non-existing function it will continue to compile and only emit a warning.
And compilation warnings are hidden by default for both `gem install` and `bundle install` (`gem install -V somegem` shows them).

A concrete example is the sqlite3 gem, if we use version 1.3.13 it fails only at runtime:
```
$ gem install sqlite3:1.3.13
Fetching sqlite3-1.3.13.gem
Building native extensions. This could take a while...
Successfully installed sqlite3-1.3.13
1 gem installed

$ ruby -rsqlite3 -e 'db = SQLite3::Database.new "test.db"; p db'
ruby: symbol lookup error: /home/eregon/.rubies/ruby-3.0.2/lib/ruby/gems/3.0.0/gems/sqlite3-1.3.13/lib/sqlite3/sqlite3_native.so: undefined symbol: rb_check_safe_obj
```
This is not nice, it should have failed clearly at compile time, saying the function does not exist.

There is a compiler warning, which can only be seen with (and so most users would miss it):
```
$ gem install -V sqlite3:1.3.13
...
database.c: In function ‘initialize’:
database.c:60:3: warning: implicit declaration of function ‘rb_check_safe_obj’; did you mean ‘rb_check_safe_str’? [-Wimplicit-function-declaration]
   60 |   rb_check_safe_obj(file);
      |   ^~~~~~~~~~~~~~~~~
      |   rb_check_safe_str
...
```

Also multiple CRuby releases are broken on macOS which seems to enable `-Werror=implicit-function-declaration` by default (e.g., #17777).
EDIT: `-Werror=implicit-function-declaration` is now default for building CRuby, but not for C extensions.

How about we just always enable `-Werror=implicit-function-declaration` for all C extensions? (builtin or not). 

It:
* shows clear errors early on and shows where the missing function is called (explained just above), instead of delaying them to runtime
* never compiles a call in C with the wrong type due to a missing include

From https://github.com/oracle/truffleruby/issues/2618



-- 
https://bugs.ruby-lang.org/

Unsubscribe: <mailto:ruby-core-request@ruby-lang.org?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>

In This Thread