Re: [boost] Usability of http://boost.sourceforge.net/regression-logs

From: Beman Dawes (bdawes_at_[hidden])
Date: 2003-01-30 15:16:54

At 11:29 AM 1/30/2003, Douglas Gregor wrote:

 >On Thursday 30 January 2003 11:05 am, David Abrahams wrote:
 >> > [Yes, I'm beginning to sound like a broken record, but I really
 >> > don't think we should ever be generating documentation directly from
 >> > C++ code.]
 >>
 >> I can't see any relevance.  Care to explain?
 >
 >I stated that _very_ poorly. I meant that I don't believe we should be
 >writing HTML directly: not by ourselves for documentation, and not with
 >C++ code for regression tests. As soon as we put data into HTML, we've
 >lost most of the structure of the data. With documentation, losing
 >structure means that it is harder to cross-reference information or
 >pick out the most important information for a particular task; with
 >the regression tests, we lose the ability to easily generate other kinds
 >of regression test tables because all of the structure of the data was
 >thrown out when the C++ program ended. I also find tweaking the HTML
 >output from C++ code to be an unnecessarily time-consuming process.

We already generate XML for the regression tests. That's what 
process_jam_log does; it produces no HTML. Bill Kempf suggested the change 
to XML last spring, and it has been a great success.

Right now we're using a C++ program (compiler_status) to convert the .xml 
files to HTML, because it is very simple and only uses tools the people who 
run regression tests already have available. The HTML it generates rarely 
changes.

The program already has options to produce one table per compiler, it that 
is desired.

--Beman