I do not think we can change the mapping between string as integer weights as we take both as input.

I am mostly 👎 on no longer picking up fonts that we used to pick up (even if the weights are a bit off).

#7931 is likely related, but does not fully fix this problem?

#7931 is likely related, but does not fully fix this problem?

I tested it using HEAD (a9a8634) with Fira Sans and each run I seem to get a random choice of:

/usr/share/fonts/TTF/FiraSans-Book.ttf
/usr/share/fonts/TTF/FiraSans-Eight.ttf
/usr/share/fonts/TTF/FiraSans-Four.ttf
/usr/share/fonts/TTF/FiraSans-Hair.ttf
/usr/share/fonts/TTF/FiraSans-Regular.ttf
/usr/share/fonts/TTF/FiraSans-Thin.ttf
/usr/share/fonts/TTF/FiraSans-Two.ttf
/usr/share/fonts/TTF/FiraSans-Ultra.ttf

Tested using:

# while rm -fr ~/.cache/matplotlib && python foo.py; do :; done
import matplotlib
matplotlib.use("Agg")
matplotlib.rcParams["font.sans-serif"] = "Fira Sans"
matplotlib.rcParams["font.weight"] = "regular"
import matplotlib.pyplot as plt
print(matplotlib.font_manager.findfont("Fira Sans"))
plt.plot([1, 2], [3, 4])
plt.savefig("foo.png")

I do not think we can change the mapping between string as integer weights as we take both as input.

OK I've restored them and pushed "Thin" down to 50 instead of 100.

I am mostly 👎 on no longer picking up fonts that we used to pick up (even if the weights are a bit off).

Perhaps we could deprioritize them and mark their weights as “uncertain”. They should still get picked up as if their weight is Regular, but if a “definitely” Regular font is found, then it takes precedence.

I tested it using HEAD (a9a8634) with Fira Sans and each run I seem to get a random choice of:

@Rufflewind I could not reproduce your results.
On my machine, your test script consistently prints /Users/kshramt/.fonts/FiraSans-Regular.otf.

I re-reproduced my results on a clean Arch container:

• Python 3.6.1
• Matplotlib 2.0.0+4221.ge232afe73
• Fira Sans 4.202 (from the .tar.gz)

(venv) [rf@arch-container matplotlib]$ while rm -fr ~/.cache/matplotlib && python foo.py; do :; done
/home/rf/.fonts/FiraSans-Eight.otf
/home/rf/.fonts/FiraSans-Book.otf
/home/rf/.fonts/FiraSans-Eight.otf
/home/rf/.fonts/FiraSans-Two.otf
/home/rf/.fonts/FiraSans-Four.otf
/home/rf/.fonts/FiraSans-Hair.otf
/home/rf/.fonts/FiraSans-Four.otf
/home/rf/.fonts/FiraSans-Hair.otf
/home/rf/.fonts/FiraSans-Book.otf
/home/rf/.fonts/FiraSans-Regular.otf
/home/rf/.fonts/FiraSans-Thin.otf
/home/rf/.fonts/FiraSans-Eight.otf
/home/rf/.fonts/FiraSans-Book.otf
/home/rf/.fonts/FiraSans-Eight.otf
/home/rf/.fonts/FiraSans-Four.otf
…

@Rufflewind Thank you for re-reproducing the results.
I could reproduce your results with FiraSans font files from https://github.com/mozilla/Fira/archive/4.202.tar.gz.
You can reproduce my results in the previous comment with FiraSans font files from https://www.fontsquirrel.com/fonts/download/fira-sans (zip file).

fc-scan ~/.fonts/FiraSans-Regular.otf shows that fullname field have changed from "Fira Sans Regular" to "Fira Sans".
According to Name IDs section of https://www.microsoft.com/typography/otspec/name.htm,

If string 2 is "Regular", it is sometimes omitted from name ID 4.

Ref: #8587

'powercycled' PR to re-trigger CI

xref #5574

I made three modification in #8607. In my opinion, the first and third are not improving, but bug fix.

Superseded by #16203; thanks for the pr.