On Saturday, 01 April 2017 at 01:01:15 pm -0400, Henry So Jr. wrote:
> On Saturday, 01 April 2017 at 11:54:55 am -0400, Br. Samuel Springuel wrote:
> > I think the bug in #1319 should be solved, but I'm not sure the proposed
> > fix is the right one: it changes code which long predates the
> > introduction of the bug.
> >
> > I've managed to narrow the range of the bug introduction to between
> > 3.0.3 and 4.0.0 but haven't had time to bisect the repository to get
> > more specific than that. I suspect it's the adjustable line height
> > code, but absent the bisecting, I can't confirm that yet.
> >
> > Part of the problem on this issue also is that the test repository
> > algorithms don't detect the difference between the correct result and
> > the wrong one on my system, even at abnormally high DPI (I tested at
> > 1200). This means that I can't even tell what the knock on effects of
> > the proposed fix are without manually inspecting every test in the
> > repository (a time prohibitive task at this point).
>
> If it's unlikely to be in the C code (which seems to be the case), and
> it already exists in the TeX Live 2016 version, then maybe we should
> consider releasing anyway, with this as a known issue. We can release
> fixes to the TeX and Lua code after TeX Live 2017 is built.
>
> What do you think?
One other thing which might help with the difference testing: you can
try changing the compare metric (in harness.sh) to AE, as that is less
forgiving.
Change ...
# trick to do floating point-like comparison in bash
metric=$(compare -metric NCC \
"$name" "$expected" null: 2>&1)
metric=$(printf '%.3f' "$metric")
metric=${metric/./}
if (( 10#$metric < 999 ))
#if ! compare -metric AE -fuzz 90% "$name" \
# "$IMAGE_CACHE/$indir/$outdir/$name" "diff-$name" 2>/dev/null
then
... to ...
if ! compare -metric AE "$name" "$expected" null: 2>/dev/null
then
If you do this, run it first on a "clean" set of tests and accept all
changes it detects to establish a new baseline.
Then, if even a single pixel is different, the test will fail.
Henry
_______________________________________________
Gregorio-devel mailing list
[email protected]
https://mail.gna.org/listinfo/gregorio-devel
