Re: makeinfo 5.0 is 34x slower when building Emacs info

[Karl Berry]

    but 224 seconds with texinfo 5.0

Paul - of course we were greatly concerned about the slowdown.  That's
why I specifically mentioned it in the announcement and NEWS file and
documentation.

The short answer is no, nothing can be done.

If/Since you edit Texinfo files often enough that it's so painful, I can
only suggest using the C makeinfo during such real-time development and
perhaps running the Perl makeinfo overnight to make sure discrepancies
haven't crept in.  Obviously that won't work forever, but I'm under the
impression that Emacs won't want to make use of Texinfo 5.0 features for
years anyway.

The long answer is that Patrice has already expended a lot of effort
profiling and speeding up the code.  There is no obvious bottleneck to
fix now.  The currently usable Perl compiler would obviously cut the
parse time to nil, but that's not where the slowdown actually happens.
The "real" Perl-to-C compiler isn't usable and has no ETA that I know
of.

It's conceivable that some pieces of the code could be replaced by
dynamic libraries, but we'd have to replace a lot to make a substantial
speedup, and it has obvious downsides of portability, maintainability,
simplicity, etc.

If I, or someone else, had come forward to write something with
equivalent functionality in C, then we'd have it in C.  No one did and I
doubt anyone ever will.  So, given the choice between leaving makeinfo
essentially as it is forever and moving forward with new features that
people were continually requesting, I chose the latter.

Wish I had better news for you, but that's the reality.

karl