Re: help wanted reproducing reported performance regression in groff 1.24

"G. Branden Robinson" <[email protected]> Tue, 17 Mar 2026 17:44:59 -0500

Newsgroups	gmane.comp.printing.groff.general
Message-ID	<20260317224459.yeo7volurmzuwtic@illithid>

At 2026-03-17T20:16:51+0000, Deri wrote:
> Using Dimich script:-
[...]
> This script is measuring grotty's performance only.

Right.  dimich made that observation in comment #8 to ticket #68145 on
Friday.

https://savannah.gnu.org/bugs/?68145

> Your script:
> 
> find data in 68145a.*.dat 
> n       1.23.0  1.24.0 
> 1       0.09    0.12 
> 2       0.17    0.23 
> 3       0.25    0.32 
> 4       0.30    0.45 
> 5       0.37    0.56 
> 6       0.44    0.68 
> 7       0.52    0.87 
> 8       0.59    0.97 
> 9       0.67    1.18 
> 10      0.74    1.35 
> 11      0.82    1.53 
> 12      0.90    1.73 
> 13      0.95    1.98 
> 14      1.03    2.21 
> 15      1.10    2.43 
> 16      1.20    2.71 
> 17      1.24    2.92 
> 18      1.32    3.17 
> 19      1.40    3.33 
> 20      1.47    3.63

So you get these numbers too when timing the whole pipeline?

The whole pipeline (when `-Z` is not specified) _includes_ _grotty_'s
runtime, so why aren't we seeing the quadratic performance hit here,
too?  It cannot be that GNU _troff_ has seen a nearly quadratic
_improvement_ in performance from 1.23.0 to 1.24.0, offsetting the
_grotty_ slowdown.  Can it?

> This script is timing the whole pipeline. Running your script with
> extra-args set to -Z:-
> 
> find data in 68145a.*.dat
> n       1.23.0  1.24.0
> 1       0.06    0.09
> 2       0.12    0.17
> 3       0.18    0.22
> 4       0.20    0.29
> 5       0.25    0.37
> 6       0.31    0.44
> 7       0.36    0.52
> 8       0.42    0.58
> 9       0.47    0.67
> 10      0.51    0.74
> 11      0.60    0.81
> 12      0.62    0.89
> 13      0.68    0.95
> 14      0.76    1.03
> 15      0.80    1.10
> 16      0.83    1.20
> 17      0.88    1.28
> 18      0.93    1.32
> 19      0.99    1.43
> 20      1.03    1.50
> 
> So although 1.24 man.tmac is slower it is not quadratic so the real
> performance hit is in grotty.

That appears to be the consensus.

My plans for next steps are in comment #15 to the same ticket.

Regards,
Branden

signature.asc (application/pgp-signature, 833 B)