[Dev-luatex] Varying behaviour for ligatures
tex-6828 at sdu.se
Mon Sep 5 16:24:19 CEST 2016
> i can't replicate this
This is a long answer, so please bear with me.
Bizarrely, I couldn't either, at first. I ran the same .tex file
100 times, and got the same result. Not good for my self-confidence.
I then ran the test 20 times on a subset of 500 files over night.
The tests are run from a Makefile. Ten of the tests runs run with six
parallel jobs (make -j6), the other ten run strictly sequentially (make -j1).
All the tests are run once by each invocation of make.
There are a total of 20 sequential invocations of make.
This garantees that the same .tex file is not processed more than once at
any given time.
loop 10 times
Run make -j6 with a new --result directory
make makes each target
Run make -j1 with a new --result directory
make makes each target
Finally I compare consequtive runs.
Now I have som statistics gathered from the logfile directories.
Runs are numbered 11 to 20.
j1/diffs/11-12: 66 / 85 / 487
j1/diffs/12-13: 2 / 28 / 487
j1/diffs/13-14: 2 / 29 / 487
j1/diffs/14-15: 1 / 25 / 487
j1/diffs/15-16: 42 / 66 / 487
j1/diffs/16-17: 104 / 123 / 487
j1/diffs/17-18: 109 / 129 / 487
j1/diffs/18-19: 112 / 129 / 487
j6/diffs/11-12: 122 / 139 / 486
j6/diffs/12-13: 68 / 87 / 486
j6/diffs/13-14: 4 / 29 / 486
j6/diffs/14-15: 4 / 28 / 486
j6/diffs/15-16: 41 / 66 / 486
j6/diffs/16-17: 100 / 122 / 486
j6/diffs/17-18: 63 / 86 / 486
j6/diffs/18-19: 2 / 29 / 486
j6/diffs/19-20: 43 / 66 / 487
Th first line shows that there are
- 66 diff files containing the string "<006600"
- 85 non-empty diff files
- 487 diff files in total
From line 4 we see that it is possible to run 487 tests to completion,
and only get a single error.
(There are 13 files that consistently do not generate any output)
This may explain why you could not replicate the error. Try running
the same file lots of times.
The different number of diff files (486 vs 487) is because of epdf/merge-002.pdf,
which depends on epdf/merge-001.pdf, which is sometimes there, sometimes not,
depending on make order.
Running make -j6 is not a problem, something that I suspected
after my initial failure to replicate the error.
> (you can send me a test off-list)
Do you want some of my log files, or the Makefiles I use to run the tests?
(The complete set of log files are way beyond my 1 Mbit upload speed.)
More information about the dev-luatex