replace the embedded font by just replacing the whole font contents directly

[JackieFei]

Is your feature request related to a problem? Please describe.

for example : pdf already embedded full set of serif.ttf, wish to be replaced by full set of sans.ttf

Can we replace the embedded font by just replacing the whole font contents directly? (without re-writing text)

reference:
#855
https://github.com/pymupdf/PyMuPDF-Utilities/tree/master/font-replacement
(The repl-font.py is using TextWriter to re-write text)

Describe the solution you'd like

flow description:

1. use pymupdf fitz open pdf
2. get serif.ttf xref number
3. use some methods to get sans.ttf object contents
4. use some methods (+xref number) to replace serif.ttf object contents with sans.ttf object contents
5. save (sans.ttf embedded, serif.ttf dropped) new pdf

expect:
pdf all text written by serif.ttf can displayed with sans.ttf

Describe alternatives you've considered

Additional context

[JorjMcKie]

No, this is impossible - meaning it cannot ever work. Not even if you wanted to replace a sans font by some other sans font.

There are several reasons for this, here are just two:

• every single glyph has its own metrics (at least for proportional fonts), e.g. an "a" in Palatino has a different width than in Arial. In addition the relative width within a font is different from that of another font.
• Text in a PDF is encoded with a glyph number per unicode. This number is in general not equal to the unicode number and it is font-specific. So if text is not rewritten with the new font, the result would be unreadable.

The next version will however contain a new Document.subset_fonts() that replaces a font with its subset - without rewriting text.

[JorjMcKie]

Example: 3 sans fonts. Closely look at the word "Hello" and you will notice different absolute and relative, intra-font character widths:

All three use a different mapping of unicode -> glyph number:

BT
/F0 11 Tf
1 0 0 1 100 742 Tm
[<00290046004D004D0050000D000100290046004D005700460055004A004400420002>] TJ
0 G
/F1 11 Tf
1 0 0 1 100 722 Tm
[<002B0048004F004F0052000F0003003100520057005200030036004400510056000300350048004A0058004F004400550004>] TJ
0 G
/F2 11 Tf
1 0 0 1 100 702 Tm
[<007301A9021B021B02440E850003006501F6027C015A006800B9000300EC01A901D702AF021B015A027C0E89>] TJ
ET

The first 4 hex digits for example in each case represent the glyph number for "H" of "Hello". So 0x0029 = "H" in Helvetica, 0x002B in Noto Sans and 0x0073 in Fira Go.
Correspondingly, if I now do a cyclic exchange of the 3 fonts (/F0->/F1->/F2->/F0) the PDF will show you this:

[JackieFei]

Thanks for your test.

Q1: Can you make a more test with the same company's fonts, just like Noto Serif and Noto Sans, maybe the text glyph numbers are identical?

Q2 : Is the action do a cyclic exchange of the 3 fonts done by PyMuPDF functions? How to do it?

Thanks in advanced.

[JorjMcKie]

Q1: Sometimes they are, sometimes they are not. You can check that yourself: Font.has_glyph(ord("A")) returns the glyph number.
Q2: I just used an editor and changed the numbers manually.

[JorjMcKie]

Addendum for Q1: for a font supported by MuPDF, there is the array Font.valid_codepoints() which delivers the unicode number for every existing glyph id of the font. Two fonts have the same mapping if these arrays are equal obviously.
E.g. fitz.Font("tiro").valid_codepoints() == fitz.Font("helv).valid_codepoints() is True.

[JorjMcKie]

In any case, my first objection remains true: fitz.Font("helv").text_length(text) != fitz.Font("tiro").text_length(text) practically always - even though these two (and "cour" too) happen to have the same glyph mapping.

[JorjMcKie]

A final comment on this:
Version 1.18.9 will have an additional way to insert a font in a PDF. Then you can experiment a little by doing this kind of thing:

# assume 'xref1' is the xref of a font you wish to replace
# insert the desired new font:
xref2 = doc._insert_font(filename=...)
str_xref2 = doc.xref_object(xref2)  # get definition syntax of new font
doc.update_object(xref1, str_xref2)  # replace old font defintion
...
doc.save(..., garbage=3,...)  # will remove unused objects