On 2023/06/16 15:07, Leonardo Brás wrote:
> On Thu, 2023-06-15 at 19:43 +0900, Akira Yokosawa wrote:
>> [+To: Leo]
>>
>> On 2023/06/15 18:48, Akira Yokosawa wrote:
>>> Hi Paul,
>>>
>>> Commit a80a57b21a5e ("fixsvgfonts.sh: Convert sans-serif into 'DejaVu Sans'")
>>> assumed that having "DejaVu Sans Mono" is sufficent for "DejaVu Sans".
>>>
>>> It turns out that docker/Dockerfile.fedora doesn't cover "DejaVu Sans".
>>> Fedora's dejavu-sans-mono-fonts package doesn't contain "DejaVu Sans Mono"!
>>> There is a meta package of the name dejavu-fonts-all which covers them
>>> both as well as dejavu-serif-fonts.
>>>
>>> It also turns out that Answer to #9 in FAQ-BUILD.txt saying:
>>>
>>> As for Ubuntu and Fedora, packages listed in #5 should cover
>>> all the font families needed.
>>>
>>> is wrong on the Fedora front. DejaVu and Liberation fonts need
>>> font packages dejavu-fonts-all and liberation-fonts.
>>>
>>> Furthermore, Fedora 38 (released this April) has a regression of
>>> Inkscape where font markup is corrupted in SVG --> PDF conversion.
>>
>> Leo, just for you info, I see the same regression under Arch Linux.
>
> Hello Akira, thanks for letting me know!
>
>>
>> Pick a perfbook.pdf from your gitlab-ci build, and run:
>>
>> pdffonts perfbook.pdf
>>
>> You will see something like (I don't know how the corrupted encoding
>> will work in your mailbox...):
>>
>> name type encoding emb sub uni object ID
>> ------------------------------------ ----------------- ---------------- --- --- --- ---------
>> WGDPYX+LMMono8-Regular Type 1 Custom yes yes yes 1722 0
>> ULGKTM+TeXGyreTermesX-Regular Type 1 Custom yes yes yes 1725 0
>> OTFSIT+LMMono12-Regular Type 1 Custom yes yes yes 1726 0
>> IYVMBP+TeXGyreTermesX-Bold Type 1 Custom yes yes yes 1741 0
>> PSUFXP+TeXGyreTermes-Regular Type 1 Custom yes yes yes 1742 0
>> IMBSNT+NimbusRomNo9L-Regu-Slant_167 Type 1 Custom yes yes yes 1929 0
>> NWEUXR+LMMonoLt10-Bold Type 1 Custom yes yes yes 2227 0
>> EOPMEH+LinBiolinumTI Type 1 Custom yes yes yes 2673 0
>> UHVCIJ+LinBiolinumT Type 1 Custom yes yes yes 2674 0
>> EOPMEH+LinBiolinumTI Type 1 Custom yes yes yes 2677 0
>> OIRARM+LMMono10-Regular Type 1 Custom yes yes yes 2764 0
>> TLODOC+txsys Type 1 Builtin yes yes yes 2828 0
>> UFJTMD+StandardSymL Type 1 Builtin yes yes yes 2830 0
>> IGSKIX+NewTXMI5 Type 1 Builtin yes yes yes 2831 0
>> VVGNAR+TeXGyreTermesX-Italic Type 1 Custom yes yes yes 2921 0
>> YFRDMR+NimbusSanL-Regu Type 1 Custom yes yes no 2980 0
>> YFRDMR+NimbusSanL-Regu Type 1 Custom yes yes no 3012 0
>> JZAPTL+k湌顕 CID Type 0C Identity-H yes yes yes 3038 0
>> XCILRI+i湌顕 Type 1C WinAnsi yes yes yes 3039 0
>> CZRRGJ+=潌顕 CID Type 0C Identity-H yes yes yes 3040 0
>> LAEXLB+o湌顕 Type 1C WinAnsi yes yes yes 3041 0
>> WJVJMV+NimbusSans-Regular Type 1C Custom yes yes no 3066 0
>> WGDQZH+NimbusSans-Regular Type 1C WinAnsi yes yes no 3129 0
>> LSRNIT+NimbusSans-BoldItalic Type 1C WinAnsi yes yes no 3130 0
>> WGDQZH+NimbusSans-Regular Type 1C WinAnsi yes yes no 3159 0
>> LSRNIT+NimbusSans-BoldItalic Type 1C WinAnsi yes yes no 3160 0
>> DEIMBC+j솛慕 TrueType WinAnsi yes yes yes 3209 0
>> RWADWA+蛒ୖ TrueType WinAnsi yes yes yes 3238 0
>> HDCMTE+蛒ୖ TrueType WinAnsi yes yes yes 3239 0
>> JGKXJS+½���䕖 Type 1C WinAnsi yes yes yes 3269 0
>> ANXDLC+Xۨ㝖 TrueType WinAnsi yes yes yes 3320 0
>> RUPDBM+ѓ㡖 Type 1C WinAnsi yes yes yes 3553 00
>> LWHAZS+秞煕 Type 1C WinAnsi yes yes yes 3560 0
>> RQGRST+A绞煕 TrueType WinAnsi yes yes yes 3561 0
>>
>> [...]
>>
>> Expected output is:
>>
>> name type encoding emb sub uni object ID
>> ------------------------------------ ----------------- ---------------- --- --- --- ---------
>> WGDPYX+LMMono8-Regular Type 1 Custom yes yes yes 1722 0
>> ULGKTM+TeXGyreTermesX-Regular Type 1 Custom yes yes yes 1725 0
>> ICXLWS+LMMono12-Regular Type 1 Custom yes yes yes 1726 0
>> IYVMBP+TeXGyreTermesX-Bold Type 1 Custom yes yes yes 1741 0
>> PSUFXP+TeXGyreTermes-Regular Type 1 Custom yes yes yes 1742 0
>> IMBSNT+NimbusRomNo9L-Regu-Slant_167 Type 1 Custom yes yes yes 1929 0
>> NUKNCC+LMMonoLt10-Bold Type 1 Custom yes yes yes 2227 0
>> EOPMEH+LinBiolinumTI Type 1 Custom yes yes yes 2673 0
>> UHVCIJ+LinBiolinumT Type 1 Custom yes yes yes 2674 0
>> EOPMEH+LinBiolinumTI Type 1 Custom yes yes yes 2677 0
>> OIRARM+LMMono10-Regular Type 1 Custom yes yes yes 2764 0
>> TLODOC+txsys Type 1 Builtin yes yes yes 2828 0
>> UFJTMD+StandardSymL Type 1 Builtin yes yes yes 2830 0
>> IGSKIX+NewTXMI5 Type 1 Builtin yes yes yes 2831 0
>> VVGNAR+TeXGyreTermesX-Italic Type 1 Custom yes yes yes 2921 0
>> YFRDMR+NimbusSanL-Regu Type 1 Custom yes yes no 2980 0
>> YFRDMR+NimbusSanL-Regu Type 1 Custom yes yes no 3012 0
>> QNDHMI+NimbusSans-Regular CID Type 0C Identity-H yes yes yes 3038 0
>> JKBTOX+NimbusSans-Regular Type 1C WinAnsi yes yes yes 3039 0
>> BIKSFC+NimbusSans-Bold CID Type 0C Identity-H yes yes yes 3040 0
>> XFXZKY+NimbusSans-Bold Type 1C WinAnsi yes yes yes 3041 0
>> WJVJMV+NimbusSans-Regular Type 1C Custom yes yes no 3066 0
>> WGDQZH+NimbusSans-Regular Type 1C WinAnsi yes yes no 3129 0
>> LSRNIT+NimbusSans-BoldItalic Type 1C WinAnsi yes yes no 3130 0
>> WGDQZH+NimbusSans-Regular Type 1C WinAnsi yes yes no 3159 0
>> LSRNIT+NimbusSans-BoldItalic Type 1C WinAnsi yes yes no 3160 0
>> KPAYFY+steelcitycomic TrueType WinAnsi yes yes yes 3209 0
>> IGYOWD+steelcitycomic TrueType WinAnsi yes yes yes 3238 0
>> LVRVOR+steelcitycomic TrueType WinAnsi yes yes yes 3239 0
>> OCNFNQ+FreeMonoBold Type 1C WinAnsi yes yes yes 3269 0
>> DHHOAV+DejaVuSans TrueType WinAnsi yes yes yes 3320 0
>> RVTHEX+DejaVuSansMono-Bold TrueType WinAnsi yes yes yes 3479 0
>> ODJFYL+FreeMonoBold Type 1C WinAnsi yes yes yes 3553 0
>> HAXSLA+URWGothic-DemiOblique Type 1C WinAnsi yes yes yes 3560 0
>> [...]
>>
>
> IIUC, the issue is the appearing of non-ascii chars in the output of pdffonts,
> is that correct?
Right.
>
> Arch Linux provides two packages for DejaVu fonts:
> ttf-dejavu and ttf-dejavu-nerd, and both seem to provide DejaVu-Sans-mono
>
> In this test, I install them both, just for the sake of trying to fix the issue:
> https://gitlab.com/linux-kernel/perfbook/-/jobs/4485677006
>
> I also added pdffonts command in the end to make it easier to verify if
> everything is fine, which still outputs non-ascii names.
>
> When I grep for DejaVuSansMono I see them appearing in the output of the job,
> and also in a previous pdf I have downloaded.
>
> I don't quite understand this, but maybe the issue is missing other fonts?
>
> Please help me understand the issue.
As far as I see, there is no font problem on your side.
I think this is regression of Inkscape.
As a matter of fact, Inkscape packages on most rolling-release distros
are having a more annoying regression, random SIGSEGV when run from
CLI under Gnome desktop environment.
See: https://gitlab.com/inkscape/inkscape/-/issues/4177
"glib2 2.76.0 breaks Command Line export"
This issue was opened back on January 24, 2023.
It was once believed to have been resolved with a fix in gtk3, but the
issue still remains after the fixed version of gtk3 reached Fedora 38.
For gitlab-ci, where there is no Gnome desktop involved, this issue
has never affected, but the font markup corruption has sneaked in
without being reported by anyone.
I happened to test Dockerfile.fedora (from fedora:38) and checked if
"DejaVu Sans" was used in Intel_Core2_arch-simplified.pdf as intended,
and failed to see the font markup.
So, I think there is nothing wrong in your gitlab-ci.yaml script.
You need to wait until the regression is fixed in the Arch Linux
Inkscape or its dependency.
I have no idea how long that would take, though.
Or you could switch the container to a fedora:37 or ubuntu:jammy
based one for the moment.
Thanks, Akira
>
> Best regards,
> Leo
>
>
>
>>
>> Thanks, Akira
>>
>>>
>>> So Deckerfile.fedora should stay with Fedora 37 for the moment.
>>>
>>> Quick search of upstream Inkscape's issue tracker didn't hit the
>>> font markup regression. I'll open an issue there unless somebody
>>> else beats me to it.
>>>
>>> So, this patch set consists of the following
>>>
>>> Patch 1/6 adds check of "DejaVu Sans" as nice-to-have font.
>>>
>>> Patch 2/6 updates FAQ-BUILD.txt and Dockerfile.fedora for nice-to-have
>>> font packages.
>>>
>>> Patch 3/6 updates Dockerfile.fedora to stay with Fedora 37, using an ARG
>>> variable so that building from fedora:38 can be possible.
>>> It also adds popller-utils to check the font markup info in PDF properties.
>>>
>>> Patch 4/6 updates Dockerfile (from ubuntu) so that the default base
>>> image is the "latest" (jammy at the moment). There is no reason to stick
>>> with focal (20.04).
>>>
>>> Patch 5/6 adds poppler-utils to Dockerfile (from ubuntu) to detect
>>> Inkscape regression early in case it actually sneaks in.
>>> (Note: Ubuntu 23.04 is safe at the moment.)
>>>
>>> Patch 6/6 changes the default uid:gid pair of container images built
>>> from Dockerfiles to 0:0, which is for rootless mode docker/podman.
>>> Repository at dockerhub for prebuilt images is changed to
>>> akiyks/perfbook-build, where the tags "latest" and "fedora" have
>>> uid:gid pair of 0:0.
>>>
>>> Thanks, Akira
>>> --
>>> Akira Yokosawa (6):
>>> Makefile: Add 'DejaVu Sans' to nice-to-have fonts
>>> FAQ-BUILD: Update nice-to-have fonts for SVG figures
>>> docker/Dockerfile.fedora: Stay with Fedora 37 for the moment
>>> docker/Dockerfile: Use 'latest' as the default tag
>>> docker/Dockerfile: Add poppler-utils package
>>> Dockerfile: Make uid:gid = 0:0 the default
>>>
>>> FAQ-BUILD.txt | 37 ++++++++++++++++++++-----------------
>>> Makefile | 15 ++++++++++++---
>>> docker/Dockerfile | 9 +++++----
>>> docker/Dockerfile.fedora | 16 +++++++++++-----
>>> 4 files changed, 48 insertions(+), 29 deletions(-)
>>>
>>>
>>> base-commit: a80a57b21a5e8ea3dfa90471126920a1131682f7
>
