[AUCTeX] very significant slow down whenever tracing

[jfbu]

Hi,

In real life one often has to debug problems.

For this one needs to turn TeX tracing on.

99% of the time I do \tracingmacros1, here in
this minimal example I will use trace package
and its \traceon


\documentclass{article}
\usepackage{trace}
\begin{document}
\traceon
\makeatletter
\count@ 1000
\loop
\ifnum\count@ > 0
\advance\count@ -1
\repeat
\traceoff
\end{document}


when executed in a command line terminal,

$ time pdflatex testauctexwithtrace
.... many lines of tracing output ...
(./testauctexwithtrace.aux) )
(see the transcript file for additional information)
No pages of output.
Transcript written on testauctexwithtrace.log.

real    0m0.142s
user    0m0.112s
sys     0m0.020s


And with 1000 replaced by 11000, I get

$ time pdflatex testauctexwithtrace
.... many lines of tracing output ...
(./testauctexwithtrace.aux) )
(see the transcript file for additional information)
No pages of output.
Transcript written on testauctexwithtrace.log.

real    0m0.460s
user    0m0.350s
sys     0m0.094s

allowing me to estimate the real cost of the lines
in log with 1000 iterations to about (0.35-0.11)/10 = 0.024s
and this includes the time by system to print to terminal...

But I have run this from an Emacs buffer
the first one takes about 5 seconds
and the one with 10000 iterations about 50 seconds.

In real life I very often encounter situations with
about 10 seconds extra compilation time. And this,
I have to do many times in a row, sometimes. So
this builds-up to significant amount of waiting time.

Could AUCTeX parsing of log output be delayed until
after the TeX run completes ?

Any source leading to voluminous logs creates
significant slow-down compared to running the
TeX run directly in a command line shell.

This can already be felt when loading many packages
or having many font-encodings in pdflatex with utf8

Best,

Jean-François