Re: [Groff] Lack of professionalism ....

[Ingo Schwarze]

Hi,

Werner LEMBERG wrote on Wed, Mar 11, 2015 at 12:29:15AM +0100:
> Peter Schaffter wrote:
>> Doug McIlroy wrote:

>>> So it looks to me as if the policy of distributing mildly
>>> compressed macro packages has only two perceptible effects: it
>>> complicates maintenance and it complicates understanding.  I am
>>> thus led to believe that this is yet another instance of ungainly
>>> galloping gnus departing from Unix's original path of simplicity
>>> and transparency.

>> Well said.  Truth is, I've never been happy with om.tmac being
>> stripped of comments and indenting.  [...]

> It's easy to revert this, and maybe it's time to do this, given that
> computers are now much faster than a few years ago.

Here is a simple measurement with groff built from git
on my notebook (ThinkPad Z61m running OpenBSD 5.7-release/i386):

Format all 4974 OpenBSD manuals with "tbl | groff -Wall -mandoc -Tascii -P-c":
    5m46.13s real     4m49.62s user     2m31.18s system

Only the ksh(1) manual, which is the largest mdoc(7) one:
    0m1.20s real     0m1.25s user     0m0.14s system

The same with unstripped man and mdoc macros:
    7m6.01s real     6m2.61s user     2m37.46s system    [all 4974]
    0m1.36s real     0m1.50s user     0m0.10s system     [ksh only]

For comparison, the same with mandoc(1):
    0m53.76s real     0m20.41s user     0m28.73s system  [all 4974]
    0m0.05s real     0m0.05s user     0m0.00s system     [ksh only]

While groff is a blazingly fast typesetter, it is a slow documentation
formatter (though not nearly as bad as DocBook based toolchains,
which tend to be agonizingly slow).  But apparently, while the
effect of stripping macro sets is easily measureable, it doesn't
matter *that* much.  The above numbers suggest the speedup to be
of the order of 10-20% for a mixture of man(7) and mdoc(7),
which isn't very impressive given that a fast formatter does the
job in less than a sixth of the time.

Even for the ksh(1) manual, which is pure mdoc(7), the speed gain
doesn't seem more relevant than for the mixture.  I'm a bit
suprised because Werner seemed to remember bigger benefits for
mdoc(7).  Maybe somebody else can confirm with different input
files on a different platform.

So i tend to agree with Werner, Peter, and Doug that the benefit
in simplicity and clarity of installing unstripped may outweigh
the gain of speed from stripping.

Yours,
  Ingo