`cmptxtrender` identifies differences when comparing a font file with itself

[bobh0303]

In the following "regression" style test, cmptxtrender is asked to compare rendering of a font file with that of the same font file, but it reports differences:

$ cd font-scheherazade/references/4.000
$ cmptxtrender -e ot -e ot -s "arab" -l "" -t "../../tests/AllChars-auto.ftml" --outputtype json -g ../../source/glyph_data.csv "ScheherazadeNew-Regular.ttf" "ScheherazadeNew-Regular.ttf"
{
"U+060C" : [[["comma-ar.downward", 0.0, 0.0], ["_adv_", 929.0, 0.0]], [["comma-ar", 0.0, 0.0], ["_adv_", 929.0, 0.0]]],
"U+061B" : [[["semicolon-ar.downward", 0.0, 0.0], ["_adv_", 929.0, 0.0]], [["semicolon-ar", 0.0, 0.0], ["_adv_", 929.0, 0.0]]],
"":[]}
$

NB: This failure was seen in Sch project as a result of smith test which was comparing rendering of results/*.ttf with corresponding fonts in references/, so passing the same font file as the two arguments to cmptxtrender is not the issue, but is done here to make it easier to replicate and possibly debug.

[bobh0303]

Observations:

• all the false regression failures I see in smith test are tests involving font features (in the above it was cv84=1) but not all tests involving features were failures.
• Trying to replicate this with a plain-text input file and -f cv84=1 on the command-line has, so far, not been successful.

[bobh0303]

A short ftml file that demonstrates the problem .... just remove the .txt extension first.
short.ftml.txt

[nrsiward]

I've seen the same bug in Gentium. I've attached a repeatable test case.

There are more failures using the full font-gentium/tests/features.ftml file all involving the application of features but not all tests with features fail (as Bob found). The text rendered in the browser for a failed test looks correct. The glyph names and advance for variant glyphs appear as though features are not being applied.

Run cmptxtrender -t features_cv13.ftml -e ot Gentium-Regular.ttf Gentium-Regular.ttf > cmp_GR.html

cmptxtrender_bug_cv13.zip

[nrsiward]

I've also seen this bug occurring in Andika Six when comparing those built ttfs to the v6.200 ttfs.

[bobh0303]

When comparing a font with itself (by building, copying into the references folder, then running CTR_UHB=1 smith test:

• With Scheherazade, test result files are all zero length, which is exactly what is expected -- yay!
• Gentium, however, still has differences -- in some cases they are glyph-name changes which is quite weird.

(When building Gentium, I suggest using its --quick option so it builds just two fonts)