MarkdownViewer sometimes lands far from the anchor after `go("doc.md#heading")` when the table of contents is shown

Open
#6,786 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Assessment

Difficulty
4/5
Estimated time
3-5 days
Newbie friendliness
30/100
Issue type
Bug
Clarity
Mostly clear
Activity status
Active
Tech stack
python
Domain
cli

Research direction

Start with the repro script in the issue, run it with the table of contents shown and then with --no-toc to confirm the timing dependence. The relevant code is Markdown.load and goto_anchor in the Markdown widget, plus the table of contents Tree sizing on idle. Done means the heading lands at the top consistently across repeated runs, with the test suite still passing.

Written by the indexing model from the issue text.

Description

Hi! I hit this while writing generated tests against Textual's widgets.

After MarkdownViewer.go("doc.md#target") (or a link to doc.md#target), the heading usually ends up at the top of the viewer. But in a few runs out of a hundred it ends up 50 or more rows lower, off screen. It only happens with the table of contents shown.

import asyncio
import os
import sys
import tempfile
from pathlib import Path

import textual
from textual.app import App, ComposeResult
from textual.widgets import MarkdownViewer

RUNS = int(sys.argv[1]) if len(sys.argv) > 1 else 50
SHOW_TOC = "--no-toc" not in sys.argv

PARAGRAPH = " ".join(["Some words that wrap differently at different widths."] * 6)
DOCUMENT = "\n\n".join(
    f"## Section {n} with a fairly long heading title\n\n{PARAGRAPH}" for n in range(12)
) + "\n\n## Target\n\nThe target.\n\n" + "\n\n".join([PARAGRAPH] * 10)

class ViewerApp(App):
    def compose(self) -> ComposeResult:
        yield MarkdownViewer(show_table_of_contents=SHOW_TOC)

async def run_once() -> int:
    app = ViewerApp()
    async with app.run_test(size=(100, 30)) as pilot:
        viewer = app.query_one(MarkdownViewer)
        await viewer.go("doc.md#target")
        await pilot.pause(0.5)
        heading = viewer.query_one("#" + viewer.document._table_of_contents[-1][2])
        # Rows between the top of the viewer and the target heading (2 when the heading is at the top).
        return heading.region.y - viewer.region.y

os.chdir(tempfile.mkdtemp())
Path("doc.md").write_text(DOCUMENT)
print("Textual", textual.__version__)
print("table of contents:", "shown" if SHOW_TOC else "hidden")
results: dict[int, int] = {}
sequence = []
for _ in range(RUNS):
    offset = asyncio.run(run_once())
    sequence.append(offset)
    results[offset] = results.get(offset, 0) + 1
print("per run:", sequence)
for offset, count in sorted(results.items()):
    print(f"target heading {offset} rows from the top: {count} of {RUNS}")

python t17.py 200, then python t17.py 100 --no-toc, on Textual 8.2.7:

Textual 8.2.7
table of contents: shown
target heading 2 rows from the top: 190 of 200
target heading 50 rows from the top: 10 of 200
Textual 8.2.7
table of contents: hidden
target heading 2 rows from the top: 100 of 100

and on main at 5e5b7ef:

Textual 8.2.8
table of contents: shown
target heading 2 rows from the top: 191 of 200
target heading 15 rows from the top: 1 of 200
target heading 50 rows from the top: 7 of 200
target heading 146 rows from the top: 1 of 200
Textual 8.2.8
table of contents: hidden
target heading 2 rows from the top: 100 of 100

"2 rows" is the normal result (the heading's top margin). In a bad run, viewer.scroll_y is 96 while the heading's virtual_region.y is 146 and the table of contents is 51 cells wide; in a good run scroll_y is 144.

As far as I can tell, this is what happens. Markdown.load calls goto_anchor, whose scroll_visible(top=True) waits for the screen's next refresh. At the same time, the table of contents is rebuilt, and its Tree (width: auto) works out its lines on idle. If the scroll is taken before the table of contents has widened, the document is still wider and wraps less, so the heading is higher. Then the table of contents widens, the paragraphs above the heading wrap onto more lines, and the heading moves down by about 50 rows. Which comes first depends on timing (in my runs, bad results were more common while the machine was busy).

I'm not sure what the best fix is: scroll to the anchor once the layout has settled after a load, or have the viewer re-run the anchor scroll when the table of contents changes size soon after a load. I'm happy to try either if you have a preference.

Dominant language
Python
Stars
37.7k
Forks
1.5k
PR merge metrics
No merged PRs in 30d

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from Textualize/textual

All issues in Textualize/textual

Similar issues

More Python issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.