<base href> breaks fragment links

[delan]

the threads page template uses <base href>, but this breaks fragment links in posts.

for example, say you’re in /posts/1.html and you have:

<base href="/posts/">
<a href="#user-content-usb3sun-user-guide">

that becomes this, which may work but will navigate for no reason:

<a href="/posts/#user-content-usb3sun-user-guide">

[delan]

• https://stackoverflow.com/questions/1889076/what-are-the-recommendations-for-html-base-tag
• https://www.drupal.org/node/13148
• https://mademistakes.com/mastering-jekyll/site-url-baseurl/
• https://www.gatsbyjs.com/docs/how-to/previews-deploys-hosting/path-prefix/
• https://www.11ty.dev/docs/plugins/html-base/
• https://www.11ty.dev/docs/filters/url/

option 1: use js to fix fragment links from #foo to url#foo

• if done correctly, does not require navigation when clicking links
• but requires js :(

option 2: opt-in fix with a {{ |url }} filter

• easiest to understand, and the most common solution
• delegates decision about whether to fix to the author
• need to be careful to avoid triggering it in (newly) imported posts
• any syntax we use may break existing posts, needing a migration

option 3: magic fix

• requires parsing all attributes that can take urls (we already have this kinda)
• strictly speaking, requires parsing all css url() values (we don’t have this)
• requires us to decide what kinds of urls to fix, which seems tricky
  • http://host/path?query#fragment → don’t fix
  • //host/path?query#fragment → don’t fix
  • /path?query#fragment → do we fix this or not?
  • path?query#fragment → do we fix this or not?
  • ?query#fragment → how about this? (less important)
  • #fragment → don’t fix

[delan]

if we go with a magic fix, what kinds of urls do we fix? say we have base_url = "/posts/"…

if we fix this → then we link to this ↓ with	/path/to/file only	path/to/file only	both
/posts/2.html	/2.html	2.html	either
/cats/diffie	[../]../cats/diffie :( //host/cats/diffie :(	/cats/diffie	[../]../cats/diffie :( //host/cats/diffie :(
/posts/tagged/cats.html	/tagged/cats.html	tagged/cats.html	either

if we fix /path/to/file urls only…

• makes more sense, according to @ar1a and 11ty
• can’t reliably link out of base dir without also locking in a host
• breaks existing posts with attachments/... urls
• breaks existing posts that link out of base dir
• so we would need to migrate and/or reimport posts

current url → link url ↓	/posts/1.html	/posts/tagged/cats.html
2.html	/posts/2.html \o/	/posts/tagged/2.html :(
/2.html	/posts/2.html \o/	/posts/2.html \o/
/cats/diffie	/posts/cats/diffie :(	/posts/cats/diffie :(

if we fix path/to/file urls only…

• can link out of base dir with leading /
• fixes fragment links without breaking any other links

current url → link url ↓	/posts/1.html	/posts/tagged/cats.html
2.html	/posts/2.html \o/	/posts/2.html \o/
/2.html	/2.html :(	/2.html :(
/cats/diffie	/cats/diffie \o/	/cats/diffie \o/

if we fix both…

• can’t reliably link out of base dir without also locking in a host
• breaks existing posts that link out of base dir
• so we would need to migrate and/or reimport posts

current url → link url ↓	/posts/1.html	/posts/tagged/cats.html
2.html	/posts/2.html \o/	/posts/2.html \o/
/2.html	/posts/2.html \o/	/posts/2.html \o/
/cats/diffie	/posts/cats/diffie :(	/posts/cats/diffie :(

[delan]

you know your change is Good when it even “fixes” a couple of fucked up chosts :)

[delan]

fixed in d14767d0fe1ab13e01ec827cecdc4348d7331fda through f5415d8312a7e096109448e33010f6ef5bb30d31; 4e484df4221efb980741d2299bbc4ce31bb9d936 unfortunately comes with a pretty big perf regression, but i’ve managed to reclaim most of the runtime in subsequent commits:

• 1.107s feed template: add <author> and <category> for each thread (3c55b7200f6f5106a96c38704e201508f50bb34f)
• 1.131s remove reliance on <base href> in several other places (f00b1199217bb8e7e6b93cc261eb2e75994d97b8)
• 4.070s guarantee that our html output fixes relative urls (4e484df4221efb980741d2299bbc4ce31bb9d936)
• 4.057s remove stray comment in TemplatedPost::filter (f5415d8312a7e096109448e33010f6ef5bb30d31)
• 4.130s [nfc] refactor fix_relative_urls() (c07837bd9944b62755ae84edf596b02eadd0ef9d)
• 2.968s perf: don’t fix relative urls twice in threads pages (13b4e0e8089070e54a7aef7e472477afb5a36b26)
• 2.951s perf: scan only specific attributes for attachment urls (fddfb1f347b49b040981a1c82d2553cada91aa07)
• 3.113s perf: prepare to parallelise render process (f82f5dd0d048a54f9132ee73c4c223e4fe975276)
• 3.161s perf: cache ThreadsContentTemplate from individual threads (05b939fc96fdc8f7c9412b73a44cbf3225a149d6)
• 2.869s perf: don’t fix relative urls twice in threads pages (again) (68ec5783f94aa2bae41d640ba9b127710a06d986)
• 2.846s perf: reduce clones in render process (part 1) (99b9d2aebd1dfed0d789671de81718ef3f35ae26)
• 2.675s perf: reduce clones and sorts in render process (part 2) (52640efc871c1abd4985302a55bdafecad65d28c)
• 2.638s perf: reduce clones in render process (part 3) (dc4d7b1fa834920446602cd9628a768f048cca14)
• 2.723s perf: reduce clones in render process (part 4) (0b5a9121b7091cfb36d66e9c0e6b7ec558b7cd87)
• 2.673s perf: reduce clones in render process (part 5) (97706beab9a81f0d23c4175df563739551565772)
• 2.513s perf: reduce clones in render process (part 6) (0f917f72e95eaa8631985bff18c11a709671a13f)
• 2.466s perf: in render process, use HashMap when key order not needed (011a19eb03c247eddcdf698c35c93851651d4322)
• 1.171s perf: parallelise render process (ce20c153ef13a42c4d6643201d2b90feef73d62b)