New gs 10.01.0 from brew yields many subtle failures

[PaulWessel]

After doing brew update and brew upgrade I got gs 10.01.0 which seems to yield numerous tiny failures where lines are just processed and rendered slightly differently. So suddenly I have 85 failures on my hand that hours ago were fine. Hoping @GenericMappingTools/core can check to verify this. I guess 85 DVC updates are no big deal but this is kind of annoying.

[seisman]

Please note that our Linux CI is still using old gs versions. Thus, if we update the PS files, the Linux CI may break.

[PaulWessel]

As a brew master, is there a command that takes me back to 10.00?

[seisman]

There is no easy way to install old versions of a homebrew package, but you can try:

1. change into the homebrew-core directory. For me, it's /opt/homebrew/Library/Taps/homebrew/homebrew-core/Formula
2. Download the old gs recipe file ghostscript.rb (https://raw.githubusercontent.com/Homebrew/homebrew-core/7ef0570825ab93c0557c3561e66cfeb6b78686be/Formula/ghostscript.rb), which provides gs 10.0.0
3. Replace the new ghostscript.rb file with the old one
4. Reinstall ghostscript

[PaulWessel]

Step 4 redownloads 10.01 even though I had placed the previous there. Perhaps I don’t know how to “reinstall”

also, when you say you are using “old” gs, are vi talking 9.x?

[joa-quim]

Don't remember how but one can tell brew to run a locally stored formula. So you'd run brew on formula after step 2.

[PaulWessel]

Just to show some of the craziness happening once 10.01 got used (presumably also via gm). Look at this failing example 23:

Yes it still fails with what you think is not too much given the picture?

/opt/homebrew/bin/gm compare: image difference exceeds limit (0.0233046 > 0.003).

[PaulWessel]

Was able to roll back to 10.00 so failures are back to normal.

[seisman]

Step 4 redownloads 10.01 even though I had placed the previous there. Perhaps I don’t know how to “reinstall”

also, when you say you are using “old” gs, are vi talking 9.x?

This looks pretty bad. Attached are ex023 PNG images generated by gs 9.54 and gs 10.01.

A significant change is that in the gs 10.01 version, the character "k" in the contour annotations is almost invisible.

However, the PDF version from gs 10.01 seems OK.

ex23.pdf

[PaulWessel]

Yes, the PDF look fine - I wonder if this is a gm/gs issue since it makes no sense.

[seisman]

It's a gs issue because gm is not used when converting ps to png.

[KristofKoch]

Just a heads-up:

Ghostscript […] has been found vulnerable to a critical-severity remote code execution flaw.

The flaw is tracked as CVE-2023-3664, having a CVSS v3 rating of 9.8, and impacts all versions of Ghostscript before 10.01.2, which is the latest available version released three weeks ago.

Source: Bleeping Computer: Critical RCE found in popular Ghostscript open-source PDF library

[PaulWessel]

At summit we should build 10.01.2 and see if they have fixed the bugs affecting us.

[joa-quim]

Don’t think they have. In that recent report of malfunction from Dan he was using 10.01.2