stdcxx-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Martin Sebor <>
Subject Re: [VOTE] Re: regression test suite naming convention
Date Fri, 23 Mar 2007 15:44:48 GMT
Andrew Black wrote:
> Martin Sebor wrote:
>> Martin Sebor wrote:
>>> I also like option (3) best, so sounds like the home for
>>> the new regression tests is unanimously in tests/regress/
>>> The remaining question is the file name convention. Do we
>>> go with <section-number>.stdcxx-<issue-number>.cpp or do
>>> we allow things like:
>>> <section-number>.<dot-clause>.stdcxx-<issue-number>.cpp
>>> e.g., 23.vector.cons.stdcxx-123.cpp
>>> FWIW, I see no reason not to allow it since it provides
>>> additional useful detail.
>> One thing occurred to me with this convention that might be somewhat
>> of an issue: the file name can be quite long and might exceed the
>> width of the first column in the table formatted by the exec utility.
>> If exec strips characters that exceed the width of the column like
>> runall did we could either remove the "stdcxx" part of the file name
>> or modify exec to extend the width of the column to fit the longest
>> file name. Andrew, how hard do you think it would be to change exec
>> to do the latter?
> Greetings Martin.
> While it wouldn't be trivial to modify the exec utility to automatically 
> calculate the width of the first column, it would be a straightforward 
> change.  Basically, prior to running the targets, you'd need to loop 
> through the list of targets, and determine the length of the longest 
> name.  Once this value has been determined, you can use it in formatting 
> the column width.

I'm concerned about arguments that are interspersed with options,
e.g., like this:

   exec --foo short-name --bar long-file-name.

POSIX doesn't allow this but maybe our exec does as an extension?
If so, parsing this correctly would be complicated by options with
arguments, such as:

   exec --foo short-name --bar bars_argument long-file-name


> My only concern with dropping the 'stdcxx' part of the name is there 
> might be a desire to include regression tests from other bug tracking 
> systems.  However, we'd probably want to copy such incidents into Jira, 
> rendering this concern moot.
> --Andrew Black

View raw message