[tamnd]

Closes #138.

cli.md says cache_dir "follows data_dir when unset" and the config file honoured that, but --data-dir did not. kit computes its default cache dir from its default data dir before any flag is parsed, so --data-dir /somewhere/fresh moved the downloads, the Parquet, and the DuckDB file, and left the cache pointing at the old tree.

That is not only untidy. It came up while measuring #137: a run aimed at an empty directory to prove a command fails without a cache came back exit 0, because it was reading the ambient cache the whole time. A flag that looks like isolation and is not gives you the wrong answer with no sign anything was wrong.

The cache dir now follows kit's resolved data dir unless somebody named it, and "named" is the same settings.value("cache_dir") lookup the config file path already used, so CCRAWL_CACHE_DIR and a cache_dir line both pin it whichever way the data dir moved.

While in there: config show reported default as the source of cache_dir and db_path even when both had moved with the data dir. That column exists to end questions about where a value came from, and default next to a moved path starts one. Both now say derived from data_dir when that is what happened, the same as raw_dir and parquet_dir.

Measured live

Before, on the same binary with the index server down:

$ ccrawl crawls latest --data-dir /tmp/emptydata; echo $?
CC-MAIN-2026-30
0

After:

$ ccrawl cache dir
/Users/apple/data/ccrawl/cache
$ ccrawl cache dir --data-dir /tmp/dd1
/tmp/dd1/cache
$ CCRAWL_CACHE_DIR=/tmp/envcache ccrawl cache dir --data-dir /tmp/dd1
/tmp/envcache

$ ccrawl crawls latest --data-dir /tmp/emptydata; echo $?
Fetch collinfo: all 6 attempts failed ... connect: connection refused.
8
$ ccrawl crawls latest; echo $?
CC-MAIN-2026-30
0

config show, before and after:

$ ccrawl config show --data-dir /tmp/dd2 -o csv
data_dir,/tmp/dd2,flag --data-dir
cache_dir,/tmp/dd2/cache,default              <- said default, was not
db_path,/tmp/dd2/ccrawl.duckdb,default        <- same

$ ccrawl config show --data-dir /tmp/dd2 -o csv
data_dir,/tmp/dd2,flag --data-dir
cache_dir,/tmp/dd2/cache,derived from data_dir
db_path,/tmp/dd2/ccrawl.duckdb,derived from data_dir

$ CCRAWL_CACHE_DIR=/tmp/envc CCRAWL_DB_PATH=/tmp/x.duckdb ccrawl config show -o csv
cache_dir,/tmp/envc,env CCRAWL_CACHE_DIR
db_path,/tmp/x.duckdb,env CCRAWL_DB_PATH

Done when

• --data-dir moves the cache dir with it, the way the reference says it does
• a cache dir named in the config file or the environment stays where it was put
• config show names the data dir as the source of a path that followed it
• the configuration page shows the sources it now prints

Checks

gofmt -l . clean, go vet ./... clean, go test ./... passing, scripts/docs-drift.sh matches the binary, scripts/coverage.sh above every floor with cli at 53.1%, golangci-lint run ./... 0 issues.

New tests: TestTheDataDirFlagMovesTheCache covers the flag moving the cache, the flag not being given, and the two ways of pinning it, and TestConfigShowSaysWhichPathsFollowTheDataDir checks the source column through the command line. Both sit below the command line for the pinning cases, because the test harness exports CCRAWL_CACHE_DIR for every run so tests cannot read each other's caches.