web/search: suggest lang, file and case qualifiers in the repo search bar #4721

Open

tomplanche.com 4d ago

targeting master from tomplanche.com/core:web/repo-search-qualifiers

Open opened by and assigned to

tomplanche.com 4d ago targeting master from tomplanche.com/core:web/repo-search-qualifiers

web/search: suggest lang, file and case qualifiers in the repo search bar #

Closes #4719.

I was surprised that the search bar on a repository's search page (/<owner>/<repo>/search) was a plain text field while the one on /search suggests the language:, author: and sort: qualifiers and their values as you type. And its empty state tells the user to narrow the search with lang:go, file:readme or case:yes.

This PR gives the repository search bar the same behavior.

Changes #

web/src/lib/codeQuery.ts: new CODE_DIALECT for the zoekt qualifiers of the empty state, on the model of REPO_DIALECT.
- lang: suggests the same languages as language:, the popular ones first.
- file: is suggested as a key. Its value is a free path pattern, so it has no value list.
- case: suggests yes, no and auto, and marks any other value as invalid.
- A qualifier without a value is marked as invalid, as in the other dialects.
web/src/lib/repoQuery.ts: export LANGUAGE_OPTIONS so that both dialects share the list.
web/src/lib/components/search/CodeSearchView.svelte: in the repository scope, use TicketQueryInput with CODE_DIALECT in place of SearchBar. The clear button goes back to the repository search page without ?q=. The global code search (/search?type=code) keeps its plain field.
web/src/lib/codeQuery.test.ts: unit tests for the known keys, the key and value suggestions, the accepted value, and the case: validation.

The query that goes to org.tangled.temp.search.searchCode does not change. Zoekt already applies these qualifiers. This change is only in the UI.

Testing #

pnpm test for codeQuery.test.ts and repoQuery.test.ts: 11 tests pass.
pnpm check: 0 errors.
Manual test on /tangled.org/core/search: key and value suggestions, keyboard navigation (arrows, Tab, Enter, Escape), invalid value highlight, clear button, and submit.

Commits 0

Could not load commits.

Comments

v0 4d ago

oppi.li 3d ago

ah since we use zoekt for code search, and it does not support the same search grammar that we use in bobbin (we embed a parser that builds tantivy objects in there), this will need some more care, the query dialect is slightly different.

one example: zoekt does not support comma lists IIRC, so lang:Go,Rust will not work here, even though this tokenizer supports it. also, we use the language hint in other places, not lang!

web/src/lib/codeQuery.test.ts:40: case:YES shouldn't be rejected imo

i will need to do a bit more through research on where the discrepancies lie between zoekt and our bobbin's search query parser.

tomplanche.com 2d ago

Thanks ! I checked bobbin: codesearch.rs sends q to zoekt as is, so the input must follow the zoekt grammar.

I agree for the commas, the tokenizer splits on commas. I can add an option that turns off value lists in order for lang:Go,Rust to be invalid plus the input does not suggest a second value.
The test line 40 checks that problem("case", "YES") returns null so YES is accepted, not rejected ^^ but I think that zoekt only accepts lowercase so case:YES may fail with a 400 ? Should bobbin lowercase it, or should the input flag it ?

About lang vs language: afaik zoekt only knows lang:, plus the empty state of the page already says lang:go. If you want language: here too, bobbin must rewrite it to lang: before calling zoekt. Which one do you prefer?

I'll wait for your research before I update the PR. Feel free to ping me if you want me to handle part of it !

web/search: suggest lang, file and case qualifiers in the repo search bar #

Changes #

Testing #

Versions