1. Download the following PDF: http://wdc.custhelp.com/cgi-bin/wdc.cfg/php/enduser/fattach_get.php?p_sid=SOUT9CAh&p_tbl=9&p_id=1219&p_created=1088113257 2. Open this PDF in Adobe Reader 7.0 and Evince 0.4.0 and compare the text rendering. Expected results: Same text rendering. Actual results: Evince employs different fonts. Garamond and Univers Condensed are respectively replaced with Times New Roman and something else. Page 2 of the PDF is hard to read due to the tight character spacing. Additional information: Xpdf 3.01 also doesn't use Garamond either (Times New Roman instead) although the Univers Condensed seems to be displayed correctly there. Version details: Evince 0.4.0 Poppler 0.4.2 Xpdf 3.01
I have a similar problem with this file, http://www.dnv.se/mou/rapport_rfid.pdf The font used, and the narrow spacing makes it almost unreadable with poppler. It renders fine in Xpdf. I'm using poppler 0.5.4 (cairo).
Created attachment 10323 [details] Screenshot displaying the difference between fonts in poppler and xpdf
Created attachment 19918 [details] Another file with rendering problems Font appears similar to comment #1.
In particular, the letters run together and this makes the document very hard to read. It also maxes out the CPU (but not the memory) while trying to render the page (Evince bug http://bugzilla.gnome.org/show_bug.cgi?id=523372). By the way, my comment above was transferred here from a bug filed against Evince: http://bugzilla.gnome.org/show_bug.cgi?id=558180 . nshmyrnev@yandex.ru asked me to transfer it.
Can you check with a newer poppler?
How new? The problem is still present with Poppler 0.10.6, using this PDF: http://www.dnv.se/mou/rapport_rfid.pdf The screenshot I made in attachment 10323 [details] is still valid.
I guess this issue may be related with buildFcPattern() which does not put a preference for serif/sans-serif info to the fontconfig's searching pattern. I remember even non-embedded font descriptor in PDF can hold a Panose value which can give such hint. I will try to improve. If it's my misunderstanding, I'm sorry in advance.
Created attachment 38335 [details] [review] Patch to check Panose values in GlobalParams::getDisplayFont()
Created attachment 38336 [details] pdftoppm result of previous patch (38335)
Oops, it is not so simple issue. fontconfig does not provide (nor search with the consideration for) serif/sans-serif info. I had to pick such info from the font by opening the font via FreeType, load OS/2 table, check Panose values. But even if I can reflect serif/sans-serif flag in the font selection, it does not mean the best matching font would be chosen. I attached my patch to use Panose values in the font selection, and its result. You can find that the regular roman text is drawn by fixed-pitched Latin glyph in CJK fonts. It's very ugly. Worse than original result. More metric matching evaluation is needed. The question would be "why xpdf show good result? xpdf does more complicated work for better font substitution?" The answer is that xpdf's font substitution is more simple. It does not use fontconfig. It seems that xpdf prefers PS Type1, and find conventional typefaces (because the variety of PS Type1 fonts on free unix system is not so wide). I have to reconsider how to find better matching fonts.
Created attachment 38339 [details] [review] Patch to check Panose values in GlobalParams::getDisplayFont(), revision 2 The "proportional" value in Panose is also considered.
Created attachment 38340 [details] pdftoppm result of previous patch (38339) Slightly better (because fixed-width Latin in CJK font can be excluded), but the regular roman text is drawn by Myanmar3_ship.ttf!
I think further improvement should be done in fontconfig side. Or, poppler should have font substitution mapping database?
suzuki i'm happy you try to help, but please do not suggest silly things, FontConfig *is* the system font database, adding one to poppler is just wrong.
Albert, thank you for comment. My English is terribly poor, so please let me confirm: do you say that "font substitution mapping database in poppler" is silly idea? I was thinking about OpenOffice.org's font substitution database - OOo uses fontconfig, but has familyname-based substition map, too. Sorry for bothering you. -- Some accented characters are lost in my pdftoppm results. It seems that buildFcPattern() cannot detect the language of AGaramond, so we cannot force to choose fonts including them.
Yes, i say that fontconfig is a database for fonts and allows the user to have familyname-based substition rules, so there is no need for poppler to do the same.
Also, please if you know that a patch does not solve the problem, do not post it here, since it's not really useful to anyone.
OK, I stop further posting till when I summarize my patch.
> I think further improvement should be done in fontconfig side. There has been interest for some time to have fontconfig cache the panose info and permit matching against it. How long it will take — even if patches are provided — is an /interesting/ question.... (My proposal/patch to cache the psname languishes on the fc list and ’zilla w/o any comment. ☹)
I can confirm that the issue still exists in Evince 3.16.1 and Okular 0.23.2, which are the latest versions as of this writing. Tested using the OP's example PDF in comment #1 dated 2005-11-26.
-- GitLab Migration Automatic Message -- This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity. You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/poppler/poppler/issues/44.
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.