[python] how best to support colorized output? (with example colorized backtrace)

[python] how best to support colorized output? (with example colorized backtrace)

[Andrew Sutherland]

Not too long ago I hacked up the example python backtrace command to 
provide colorized output.

blog post:
http://www.visophyte.org/blog/2009/03/04/gaudily-colorized-gdb-backtraces-woo/

repo:
http://hg.mozilla.org/users/bugmail_asutherland.org/pythongdb-gaudy/

Thiago dropped by and commented, suggesting I post about it here.

While my solution was never intended to make it upstream (and should 
not), I'd be interesting in finding out how colorized output could be 
implemented that could be integrated.

Thiago mentioned that Tom had some ideas on this subject, and I'd be 
particularly interested in hearing them.

Thanks,
Andrew

Re: [python] how best to support colorized output? (with example colorized backtrace)

[Tom Tromey]

>>>>> "Andrew" == Andrew Sutherland <sombrero@alum.mit.edu> writes:

Andrew> http://www.visophyte.org/blog/2009/03/04/gaudily-colorized-gdb-backtraces-woo/

Cool!

Andrew> Thiago dropped by and commented, suggesting I post about it here.

Nice outreach, thanks Thiago.

Andrew> While my solution was never intended to make it upstream (and should
Andrew> not), I'd be interesting in finding out how colorized output could be
Andrew> implemented that could be integrated.

I think the only thing I would suggest would be to use the existing
filtering backtrace, and just reimplement colorization as a filter.
If that's not possible (I didn't dig into the details too much), let's
modify the existing Python backtrace command to make it possible :-)

Tom

Re: [python] how best to support colorized output? (with example colorized backtrace)

[Andrew Sutherland]

On 03/26/2009 05:18 PM, Tom Tromey wrote:
> I think the only thing I would suggest would be to use the existing
> filtering backtrace, and just reimplement colorization as a filter.
> If that's not possible (I didn't dig into the details too much), let's
> modify the existing Python backtrace command to make it possible :-)

Both the FilteringBacktrace class and its helper FrameWrapper class 
directly write to an output stream as text devoid of any semantic 
information.  (They are using sys.stdout, though not always explicitly.)

It is hard to decorate this output without operating at the level of a 
glorified colorizing regex.

I think something more along the line of a semantic stream (along the 
lines of ui-out.h but perhaps more extensible) might work out, but could 
be argued to be overkill.

For example, if the FrameWrapper did:
   stream.write_string(FILENAME, sal.symtab.filename)
   stream.write_text(":")
   stream.write_int(LINENUMBER, sal.line)
that would be wildly easier to colorize.  The default stream would just 
ignore the semantic info and dump things as strings.  The colorizing 
stream would have colors bound to certain types of things.  I am 
obviously glossing over where FILENAME AND LINENUMBER would come from.

Crossing into definite exciting overkill territory would be pushing the 
invocations of the pretty printers into the smart stream itself.  That 
way the stream could perhaps do something even more clever than the 
default pretty printer.  And if the stream was part of a GUI it might be 
able to establish some actionable linkage so you can right-click and 
have relevant actions, etc.

Andrew