527220 - using the "loc" parameter of error() or warning() crashes gcc 4.5

Status: RESOLVED FIXED

(Developer Infrastructure :: Source Code Analysis, defect)

Description

[Arpad Borsos [:Swatinem]]

This happens for example in the test_diag_loc.js testcase

  error("test_error", decl.loc);

This line leads to the following GCC error:

*** WARNING *** there are active plugins, do not report this as a bug unless you can reproduce it without enabling any plugins.
Event			Plugins
PLUGIN_FINISH_TYPE	gcc_dehydra 
PLUGIN_FINISH_UNIT	gcc_dehydra 
PLUGIN_CXX_CP_PRE_GENERICIZE	gcc_dehydra 
PLUGIN_FINISH	gcc_dehydra 
PLUGIN_START_UNIT	gcc_dehydra 
In function 'foo(int, float, char*)':
in pp_base_format, at pretty-print.c:546
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

The testcase runs fine when you leave out the loc parameter like this:
  error("test_error");
same applies to warning() too.

Comment 1

[Arpad Borsos [:Swatinem]]

Attachment: patch

Looks like gcc 4.5 needs error_at rather than error with a format parameter.

Comment 2

[Arpad Borsos [:Swatinem]]

While working on this bug, I also noticed that the documentation of the builtin functions is wrong.

https://developer.mozilla.org/En/Dehydra/Function_Reference#warning%28.5bcode.2c.5d_msg_.5b.2c_loc.5d%29

It says warning can take a warning code as first argument which it does not according to the actual code.

Comment 3

[(dormant account)]

Attachment: works with 4.3

How is this uglier approach to keep compat with 4.3? I don't really care about 4.2 compat as we basically broke it until someone sits down and works out all of the 4.2 kinks again, there is no point in worrying about it.

Comment 4

[(dormant account)]

Attachment: actual patch

Comment 5

[(dormant account)]

pushed http://hg.mozilla.org/rewriting-and-analysis/dehydra/rev/f4a6c776aada

The commit log is getting Arpad heavy :)