More thorough cleanup of input whitespace #151
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
There are various cases in which inline text fails to be separated by (sufficiently many) newlines from adjacent block content. A paragraph needs a blank line (two newlines) separating it from prior text, as does an underlined header; an ATX header needs a single newline separating it from prior text. A list needs at least one newline separating it from prior text, but in general two newlines (for an ordered list starting other than at 1, which will only be recognized given a blank line before). To avoid accumulation of more newlines than necessary, take care when concatenating the results of converting consecutive tags to remove redundant newlines (keeping the greater of the number ending the prior text and the number starting the subsequent text). This is thus an alternative to matthewwithanm#108 that tries to avoid the excess newline accumulation that was a concern there, as well as fixing more cases than just paragraphs, and updating tests. Fixes matthewwithanm#92 Fixes matthewwithanm#98
This fixes various issues relating to how input whitespace is handled and how wrapping handles whitespace resulting from hard line breaks. This PR uses a branch based on that for matthewwithanm#120 to avoid conflicts with the fixes and associated test changes there. My suggestion is thus first to merge matthewwithanm#120 (which fixes two open issues), then to merge the remaining changes from this PR. Wrapping paragraphs has the effect of losing all newlines including those from `<br>` tags, contrary to HTML semantics (wrapping should be a matter of pretty-printing the output; input whitespace from the HTML input should be normalized, but `<br>` should remain as a hard line break). To fix this, we need to wrap the portions of a paragraph between hard line breaks separately. For this to work, ensure that when wrapping, all input whitespace is normalized at an early stage, including turning newlines into spaces. (Only ASCII whitespace is handled this way; `\s` is not used as it's not clear Unicode whitespace should get such normalization.) When not wrapping, there is still too much input whitespace preservation. If the input contains a blank line, that ends up as a paragraph break in the output, or breaks the header formatting when appearing in a header tag, though in terms of HTML semantics such a blank line is no different from a space. In the case of an ATX header, even a single newline appearing in the output breaks the Markdown. Thus, when not wrapping, arrange for input whitespace containing at least one `\r` or `\n` to be normalized to a single newline, and in the ATX header case, normalize to a space. Fixes matthewwithanm#130 (probably, not sure exactly what the HTML input there is) Fixes matthewwithanm#88 (a related case, anyway; the actual input in matthewwithanm#88 has already been fixed)
This improves the markdownify logic for cleaning up input whitespace that has no semantic significance in HTML. This PR uses a branch based on that for matthewwithanm#150 (which in turn is based on that for matthewwithanm#120) to avoid conflicts with those fixes. The suggested order of merging is just first to merge matthewwithanm#120, then the rest of matthewwithanm#150, then the rest of this PR. Whitespace in HTML input isn't generally significant before or after block-level elements, or at the start of end of such an element other than `<pre>`. There is some limited logic in markdownify for removing it, (a) for whitespace-only nodes in conjunction with a limited list of elements (and with questionable logic that ony removes whitespace adjacent to such an element when also inside such an element) and (b) only for trailing whitespace, in certain places in relation to lists. Replace both those places with more thorough logic using a common list of block-level elements (which could be expanded more). In general, this reduces the number of unnecessary blank lines in output from markdownify (sometimes lines with just a newline, sometimes lines containing a space as well as that newline). There are open issues about cases where propagating such input whitespace to the output actually results in badly formed Markdown output (wrongly indented output), but matthewwithanm#120 (which this builds on) fixes those issues, sometimes leaving unnecessary lines with just a space on them in the output, which are dealt with fully by the present PR. There are a few testcases that are affected because they were relying on such whitespace for good output from bad HTML input that used `<p>` or `<blockquote>` inside header tags. To keep reasonable output in those cases of bad input now input whitespace adjacent to those two tags is ignored, make the `<p>` and `<blockquote>` output explicitly include leading and trailing spaces if `convert_as_inline`; such explicit spaces seem the best that can be done for such bad input. Given those fixes, all the remaining changes needed to the expectations of existing tests seem like improvements (removing useless spaces or newlines from the output).
jsm28
added a commit
to jsm28/python-markdownify
that referenced
this pull request
Oct 3, 2024
This fixes problems with the markdownify logic for indentation inside list items. This PR uses a branch building on that for matthewwithanm#120, matthewwithanm#150 and matthewwithanm#151, so those three PRs should be merged first before merging this one. There is limited logic in markdownify for handling indentation in the case of nested lists. There are two major problems with this logic: * As it's in `convert_list`, causing a list to be indented when inside another list, it does not add indentation for any other elements such as paragraphs that may be found inside list items (or `<pre>`, `<blockquote>`, etc.), so such elements are wrongly not indented and terminate the list in the output. * It uses fixed indentation of one tab. Following CommonMark, a tab in Markdown is considered equivalent to four spaces, which is not sufficient indentation in ordered list items with a number of three or more digits. Fix both of these issues by making `convert_li` handle indentation for the contents of `<li>`, based on the length of the list item marker, rather than doing it in `convert_list` at all.
|
Having this merged would be really useful. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This improves the markdownify logic for cleaning up input whitespace
that has no semantic significance in HTML.
This PR uses a branch based on that for #150 (which in turn is based
on that for #120) to avoid conflicts with those fixes. The suggested
order of merging is just first to merge #120, then the rest of #150,
then the rest of this PR.
Whitespace in HTML input isn't generally significant before or after
block-level elements, or at the start of end of such an element other
than
<pre>. There is some limited logic in markdownify for removingit, (a) for whitespace-only nodes in conjunction with a limited list
of elements (and with questionable logic that ony removes whitespace
adjacent to such an element when also inside such an element) and (b)
only for trailing whitespace, in certain places in relation to lists.
Replace both those places with more thorough logic using a common list
of block-level elements (which could be expanded more).
In general, this reduces the number of unnecessary blank lines in
output from markdownify (sometimes lines with just a newline,
sometimes lines containing a space as well as that newline). There
are open issues about cases where propagating such input whitespace to
the output actually results in badly formed Markdown output (wrongly
indented output), but #120 (which this builds on) fixes those issues,
sometimes leaving unnecessary lines with just a space on them in the
output, which are dealt with fully by the present PR.
There are a few testcases that are affected because they were relying
on such whitespace for good output from bad HTML input that used
<p>or
<blockquote>inside header tags. To keep reasonable output inthose cases of bad input now input whitespace adjacent to those two
tags is ignored, make the
<p>and<blockquote>output explicitlyinclude leading and trailing spaces if
convert_as_inline; suchexplicit spaces seem the best that can be done for such bad input.
Given those fixes, all the remaining changes needed to the
expectations of existing tests seem like improvements (removing
useless spaces or newlines from the output).