Improve filename detection for attachments

[liZe]

Fix #1862.

[wtschueller]

I modified the new test into

import weasyprint
import base64

data = base64.encodebytes(b'Weasyprint!\n').decode("utf-8")

html = f'''
  <title>Test document</title>
  <meta charset="utf-8">
  <a
    rel="attachment"
    href="data:text/plain;base64,{data}"
    download>A link that lets you download an attachment</a>
'''

weasyprint.HTML(string=html).write_pdf('test_data_url_with_line_break.pdf') 

Then there is no attachment. I think it is the line break at the end of data resulting in the following html

  <title>Test document</title>
  <meta charset="utf-8">
  <a
    rel="attachment"
    href="data:text/plain;base64,V2Vhc3lwcmludCEK
"
    download>A link that lets you download an attachment</a>

This works with 69.0.

[liZe]

I think it is the line break at the end of data resulting in the following html

Oh, you’re right, thanks for the report. That’s because of other, unrelated changes in current main branch.

[liZe]

@wtschueller I updated the main branch and this pull request to handle these spaces correctly and test this case to avoid regressions. Everything should work now, thanks a lot for the report!

[wtschueller]

I think there is still some odd behaviour:

import weasyprint
import base64

data = base64.encodebytes(b'Weasyprint!\n'*5).decode("utf-8")

html = f'''
  <title>Test document</title>
  <meta charset="utf-8">
  <a
    rel="attachment"
    href="data:text/plain;base64,{data}"
    download>A link that lets you download an attachment</a>
'''

weasyprint.HTML(string=html).write_pdf('test_data_url_with_inner_line_break.pdf')

produces an inner line break

  <title>Test document</title>
  <meta charset="utf-8">
  <a
    rel="attachment"
    href="data:text/plain;base64,V2Vhc3lwcmludCEKV2Vhc3lwcmludCEKV2Vhc3lwcmludCEKV2Vhc3lwcmludCEKV2Vhc3lwcmlu
dCEK
"
    download>A link that lets you download an attachment</a>

and then there is still no attachment for download. If I use the same procedure for <img src=".."> then everytthing works as expected.

[liZe]

I learned so many details thanks to this example, thanks a lot…

In href attributes we have valid URL potentially surrounded by spaces. Then, let’s remove the leading and trailing spaces and it will work, right?

No.

From the URL specification:

2. If input contains any ASCII tab or newline, invalid-URL-unit validation error.
3. Remove all ASCII tab or newline from input.

So, if a URL contains a newline, it should be an error, but we should remove the newline to give this poor URL a chance to be valid.

It works in WeasyPrint for non-href attributes (href puts the link in CSS, that’s a hack that fails for all newlines), but only for base64 URLs, because Python’s function allows newlines in the base64 strings.

I’ll open a new issue to allow URLs to be surrounded by spaces and to ignore ASCII tab and newline characters in them.

[liZe]

Merging, as the remaining problems are in #2851, to be solved before next version.

Thanks again for testing and reporting!