A table.include.list entry that matches no table is raised, with the corrected entry when the mistake is a missing schema or a shell glob

[avison9]

What it changes

A new rule, captured-table-missing (warning): each table.include.list entry (or the old table.whitelist spelling) is checked against the tables in the migrations, and an entry that matches none is raised on the connector file.

Two shapes get the corrected entry as the fix, because they are how people get it wrong in practice and both follow from Debezium's documented matching (each entry is a regular expression matched against the whole schema.table name, never a substring):

warning captured-table-missing include-table-no-schema/connector.json
  ipaddrs in table.include.list matches no table in the source schema
  Debezium matches each entry against the whole schema.table name
  fix: write myschema\.ipaddrs, or remove it

warning captured-table-missing include-table-glob/connector.json
  public.bg_* in table.include.list matches no table in the source schema
  Debezium reads each entry as a regular expression, where * repeats the character before it
  fix: write public\.bg_.* to capture public.bg_items and public.bg_orders, or remove it

Anything else gets the plain fix: remove it, or check the spelling against the migrations.

Why a warning, not an error

Debezium itself only logs a warning here and keeps running. debezium/dbz#872 asks it to fail; a maintainer answered that a table may be created later, which is a legitimate case a linter reading migrations cannot rule out. When a sink actually reads the table, the existing sink-table-not-captured error still fails the pull request, as all three new corpus entries show.

What was rejected

• Checking exclude-list entries: one that matches nothing excludes nothing, the same reasoning as captured-column-missing.
• Reading an entry that already contains .* as a glob: it was written as a regular expression, and rewriting it would be a guess.
• Suggesting a nearest name for plain typos: an edit-distance guess could point at the wrong table; the plain fix is honest.

Verified

• Corpus first: include-table-no-schema (reduced from Stack Overflow 74103659, 6.9k views), include-table-glob (Stack Overflow 51345636, 5.9k views), include-table-typo. The expectations were written by hand and failed before the code; they pass as written. Every existing corpus entry is unchanged (no -update).
• gofmt -l . clean, go vet ./..., go build ./..., go test ./... pass.
• RefuseRadar at 1986043: 0 error(s), 0 warning(s), 159 info, unchanged, so no false positive on a real pipeline.
• A probe on a throwaway copy of RefuseRadar's connector with public.reports changed to reports: the rule raises reports ... matches no table, suggests public\.reports, and the existing error shows the consequence in the real sink.

Docs

README: the rule in the rules table (marked next until it is released), and the symptom rows for table.include.list not working and a table-list typo now name it. corpus/README: the three entries.

Next for you

Merge when happy. It ships in the next release tag.