Avoid invalid PDF operators when drawing SVG text

[rianmcguire]

This was causing display problems and warning messages in some PDF software.

Example HTML

<p>SVG with text:</p>
<svg width="300" height="50">
    <rect width="300" height="50" fill="none" stroke="red" />
    <text rotate="10" x="10" y="25" letter-spacing="3" font-size="30" font-family="Noto Sans">This is text</text>
</svg>

Before

PDF X-Change Viewer[1] doesn't render SVG text:

Adobe Acrobat Reader DC [2] renders SVG text:

but displays this warning message after printing (printing to a virtual printer is sufficient):

After

PDF X-Change Viewer renders SVG text:

Adobe Acrobat Reader DC now prints with no warning messages.

Root cause

I managed to narrow this down to the PDF content stream that's being emitted for the SVG text. I used pikepdf to parse it.

SVG text PDF content stream before (invalid operators marked with **):

BT []
**q []
**m [10, 25]
**cm [1, 0, 0, 1, 10, 25]
**cm [Decimal('0.984808'), Decimal('0.173648'), Decimal('-0.173648'), Decimal('0.984808'), 0, 0]
**cm [1, 0, 0, 1, -10, -25]
rg [0, 0, 0]
gs [pikepdf.Name("/a1.0")]
w [1]
J [0]
j [0]
M [4]
Tr [0]
Tm [30, 0, 0, -30, 10, 25]
Tf [pikepdf.Name("/JDIZPJ"), 1]
TJ [pikepdf.Array([ "7" ])]
**Q []
ET []

According to the PDF specification, "special graphics state" operators (q, Q, cm) and "path construction" operators (m) aren't permitted while a text object is being drawn. See page 134-135, Table 4.1 Operator categories, and Figure 4.1 Graphics objects.

PDF readers seem to be quite tolerant of invalid content streams in general, but apparently this was too much!

Content stream after:

BT []
rg [0, 0, 0]
gs [pikepdf.Name("/a1.0")]
w [1]
J [0]
j [0]
M [4]
Tr [0]
Tm [Decimal('29.544233'), Decimal('5.209445'), Decimal('5.209445'), Decimal('-29.544233'), 10, 25]
Tf [pikepdf.Name("/BBQVNM"), 1]
TJ [pikepdf.Array([ "7" ])]
ET []

[1] PDF-XChange Viewer 2.5 (Build 322.10) - https://www.tracker-software.com/product/pdf-xchange-viewer
[2] Adobe Acrobat Reader DC 2021.011.20039

[liZe]

Hello!

Thanks a lot for this pull request ❤️. There’s definitely a problem with the old code, and this PR does what needs to be done, that’s perfect!

Strange to see Adobe Reader complaining, while Ghostscript doesn’t. That’s annoying to add a non-regression test.