Repository navigation
Releases
Releases · alias2k/flusso
flusso-stream-v0.18.0
Added
- (engine) write several lane items at once per sink (#386 stage 2)
Other
- Batch.walked and Request::Backfill.after (touches all constructors)
- a published batch carries its charge until its last copy is dropped
- one lane per (index, sink): stream, Progress, sink engine, publish, requests, fence
- READMEs and a doc link no longer name the stream's watermark
- the engines confirm and fence from one Progress; the stream keeps no watermark
- Position carries its seq and the source's stamps (touches all constructors)
flusso-stream-channel-v0.18.0
Added
- (engine) write several lane items at once per sink (#386 stage 2)
Fixed
- (daemon) reset an engine's restart backoff after a run that made progress (#395 review)
- (sinks) resend items OpenSearch refuses as busy, and release a lane delivery by its ticket (#395)
Other
- the default memory budget goes back to 512 MiB
- the default memory budget is 64 MiB, not 512
- a deprecated [stream] capacity or --queue-capacity validates, warns once and is ignored
- Batch.walked and Request::Backfill.after (touches all constructors)
- a configured lane holds the backstop, not the backstop times the sinks (review 11); search.rs doc order (review 9)
- [stream] capacity is deprecated: accepted, warned about, ignored; lanes keep a 1,048,576-item backstop
- a published batch carries its charge until its last copy is dropped
- one lane per (index, sink): stream, Progress, sink engine, publish, requests, fence
- READMEs and a doc link no longer name the stream's watermark
- the engines confirm and fence from one Progress; the stream keeps no watermark
- Position carries its seq and the source's stamps (touches all constructors)
- (stream-channel) is_empty reads no lock; an acknowledged handle takes none on drop (#408)
- the sink's real bound, the measured numbers, the catch-up refresh cost and the API changes (#408 review 1 findings 3-4, review 2 findings 1, 3, 5, 6)
- concurrent writes keep per-document order, the watermark and live-first (#386 stage 2)
- an item refused as busy is resent, and a release redelivers only its ticket (#395)
flusso-source-v0.18.0
Added
- (engine) stage a sink's indexes concurrently, and validate indexes concurrently
- (engine,daemon,source-postgres) [breaking] a stop drains what is in flight before the run ends
- (source) warn on a timestamp format that refuses the ISO form
- (source) [breaking] validate array columns against the catalog (#276)
- (kernel) a column may hold an array of its type (#276)
Fixed
- (engine) a snapshot row is its own document, not a change to resolve
- (source-postgres) [breaking] emit a change only once its transaction is visible
- (source) say what an unreadable root key type costs now
- (source-postgres) Postgres decides which row answers which requested key
- (source-postgres) statements survive a column type change
- (source-postgres) [breaking] write every numeric digit as a JSON number
- (source) a check the database would not let run is reported, not dropped (#324)
- (source) [breaking] a schema that can never index a document stops the start (#316, #309)
- (source-postgres) say a lossy ids key may be stored altered; check prints a skipped probe (#299, #313)
- (source-postgres) probe a type pair once, in one round trip; pin exact ids pairs (#299, #313)
- (source-postgres) check reports a link or ids element the key can't take (#299, #312, #313)
- (source) a camelCase format name reads the ISO form; the date table follows the T form (#304)
- (source-postgres) resolve each change as it is buffered again (#295)
- (source-postgres) read a batch's root keys back in one query per index (#295)
- (source-postgres) read every key type the same way in backfill, WAL and build
- (design,source) a name applies when the press that left its box ends; key lists by identity
- (source,design) duplicate-name outcomes for constants and binary; one-line node header; palette returns focus
- (design) refuse a field name its level already has; warn about duplicates; read typed table and column names as flusso does
- (source) a domain over an array or jsonb holds an array
- review findings on array columns (#276)
- (source-postgres) a transform probe the role cannot run is not a verdict
- (source) [breaking] settle the root-key kinds; probe transforms without a table
- (source) report four schemas check accepted that cannot work
Other
- a mark records the comparison operator the walk used, so one taken with the bare
<is never resumed with the qualified one; compare_keys' doc says an unplaced lane starts from the first key - the source marks a walk's point and resumes after it in its own key order (walk_mark, resume_after, compare_keys)
- review follow-ups: no watermark left, wal_idle's half named, the argument per lane
- a walk ends at the greatest key it found, read through the key index's own operators
- the engines confirm and fence from one Progress; the stream keeps no watermark
- the source confirms and fences from Landed; the Postgres capture keeps no map
- a backfill pages each index's root keys on the builder pool
- (engine) snapshot rows are not resolved, and what a faster backfill holds in memory
- (source) a check the database would not let run is reported (#324)
- (source) a camelCase format name is a name, not a literal pattern (#304)
- warn on a timestamp format that refuses the ISO form; pin tuned mappings
flusso-source-postgres-v0.18.0
Added
- (engine,daemon,source-postgres) [breaking] a stop drains what is in flight before the run ends
- (bench) review-ab --digest and --plans over the cases roots (#347)
- (bench) review-ab shows each side's range; a wide composite-key root and a never-analysed root
- (bench) just review-ab compares a branch with its base on the Docker-backed benches
- (source) warn on a timestamp format that refuses the ISO form
- (design) author array columns (#276)
- (source) [breaking] validate array columns against the catalog (#276)
- (source-postgres) clean up an array column's value as the document is decoded (#276)
- (source-postgres) per-element transforms on an array column (#276)
- (kernel) a column may hold an array of its type (#276)
Fixed
- (engine) a snapshot row is its own document, not a change to resolve
- (engine,source-postgres) a stop that can't confirm its final position fails; a deferred sink refreshes past skipped batches
- (source-postgres) a stop right after the start doesn't wait for a status update
- (cli) bound the stop, honour a second signal and SIGHUP, follow the parent
- (script,source-postgres) a declared float in --input is a float; jsonb zero padding is not digits
- (source-postgres) a float reads as Postgres casts to float4; exact reads pinned in flusso-query
- (source-postgres) an untyped number is a decimal to a script at every magnitude
- (source-postgres) an ids of float keys reads each key as an f32
- (source-postgres) read and write every number exactly
- (source-postgres) date a snapshot by its read, not by when the stream took it
- (source-postgres) a snapshot task that can't miss a commit; bounded reads; one hold clock
- (source-postgres) [breaking] emit a change only once its transaction is visible
- (bench) a plan check that cannot read its input is a setup error (#347)
- (bench) the review harness is a bench target of its own (#347)
- (test) the guard rejects the postgres import, not call shapes
- (test) every test and bench starts Postgres through dev/test-postgres
- (source-postgres) [breaking] a filter value is never cut or rounded to its column's size
- (bench) review-ab only adopts a pre-marker cache with a real stamp and nothing foreign; --clean-results apart from --clean
- (bench) review-ab's spread narrows with more rounds; --clean deletes only its own cache; owner-labelled containers; cargo config in the stamp
- (bench) review-ab re-runs a flagged case before calling it slower, alternates sides, and runs one at a time
- (source-postgres) a batch whose prepared statement changed result type reruns once
- (source-postgres) write a batch's keys as an array constructor, padded to a power of two
- (source-postgres) a shared key whose document reads only its key orders nothing
- (source-postgres) choose a shared key's row by its primary key or the columns the document reads
- (source-postgres) a batch by key reruns by position when its cached key type is stale
- (source-postgres) a name key pairs by position, since its cast cuts a requested key
- (source-postgres) pair a batch's positions in a full join, which never runs as a nested loop
- (source-postgres) the source's own sessions run without JIT
- (source-postgres) build the batch at the root's own level and pair positions one level out
- (source-postgres) name the batched build's key list so no table's column meets it
- (source-postgres) a batched build's root probe returns one row, so it is never JIT-compiled
- (source) say what an unreadable root key type costs now
- (source-postgres) read a key type's name as the qualified identifier it is
- (source-postgres) a role that may not run has_schema_privilege still reads its key types
- (source-postgres) Postgres decides which row answers which requested key
- (source-postgres) mirror real, boolean, time and uuid casts in the batched build's matching
- (source-postgres) a key of a domain over a type the role may not use keeps indexing
- (source-postgres) the batched build serves the ids a single build serves, across key kinds
- (source-postgres) a numeric key equal to its row's is built, not tombstoned
- (source-postgres) a key is never cast to a size, and a resize refreshes the cached types
- (source-postgres) statements survive a column type change
- (source-postgres) only a declared decimal changes what the sink receives
- (source-postgres) [breaking] write every numeric digit as a JSON number
- (source-postgres) a refused link probe is cached like a verdict (#324)
- (source) a check the database would not let run is reported, not dropped (#324)
- (source) [breaking] a schema that can never index a document stops the start (#316, #309)
- (source-postgres) a Relation that omits a column is no evidence its type changed (#298)
- (source-postgres) a table's Relation message drops its cached column types (#298)
- (source-postgres) a failed document query forgets its tables' column types (#298)
- (source-postgres) wrap an array of a bytea domain; keep the session SET as a backstop (#298)
- (source-postgres) a binary field holds the column's bytes whatever bytea_output is (#298)
- (source-postgres) say a lossy ids key may be stored altered; check prints a skipped probe (#299, #313)
- (source-postgres) probe a type pair once, in one round trip; pin exact ids pairs (#299, #313)
- (source-postgres) check reports a link or ids element the key can't take (#299, #312, #313)
- (query-derive) refuse by name only what can't be a string type of your own
- (query-derive) say only what the derive promises; make the boundary test cheap
- (source-postgres) resolve each change as it is buffered again (#295)
- (source-postgres) compare a resolved value as the join compares its column (#295)
- (source-postgres) parse years past 9999, so their ids keep main's form (#295)
- (source-postgres) read a batch's root keys back in one query per index (#295)
- (source-postgres) set no startup options, so PgBouncer passes (#295)
- (source-postgres) a link names the root's own document, however it spells the key (#295)
- (source-postgres) read a key back as the text the WAL sends (#295)
- (source-postgres) refuse a jsonb or json root key (#295)
- (source-postgres) read every key type the same way in backfill, WAL and build
- (source-postgres) read bytea as hex whatever the database's bytea_output
- (source-postgres) a constant builds the value the schema holds
- (source-postgres) index a binary column as its bytes, not Postgres hex text
- (cli) script run reports a document that fails to build as the failure
- (source) a domain over an array or jsonb holds an array
- review findings on array columns (#276)
- (source-postgres) a transform probe the role cannot run is not a verdict
- (source) [breaking] settle the root-key kinds; probe transforms without a table
- (source) report four schemas check accepted that cannot work
Other
- comments say what the code does, not which review or stage changed it (review L4)
- a mark records the comparison operator the walk used, so one taken with the bare
<is never resumed with the qualified one; compare_keys' doc says an unplaced lane starts from the first key - the source marks a walk's point and resumes after it in its own key order (walk_mark, resume_after, compare_keys)
- changes are resolved on a connection of their own (step 2)
- review follow-ups: no watermark left, wal_idle's half named, the argument per lane
- a walk ends at the greatest key it found, read through the key index's own operators
- READMEs and a doc link no longer name the stream's watermark
- rustfmt the review harness
- wal_idle confirms Idle once a second, as the engine's tick does
- the source confirms and fences from Landed; the Postgres capture keeps no map
- Position carries its seq and the source's stamps (touches all constructors)
- a backfill pages each index's root keys on the builder pool
- (source-postgres) an untyped whole-number double is the integer it reads as
- (source-postgres) name the untyped number rule the reader follows
- (source-postgres) a commit arriving as the snapshot task stops is still checked
- (source-postgres) keep a snapshot fresh while transactions await a check
- (source-postgres) read the visibility snapshot on its own task
- (source-postgres) measure a small batch's plan on one connection
- (source-postgres) a small batch's plan test refuses a server that counts no plans
- (source-postgres) a small batch's cached plan is measured the same way on every major
- trailing spaces on varchar(n), an array of a sized domain, rewrapped lines
- (source-postgres) a fitting filter value still matches its row for every sized type
- (source-postgres) a f...
flusso-sink-v0.18.0
Added
- (engine) [breaking] stage timings account for each task's whole time (#399 review)
- per-stage timings in /metrics and the bench's attribution table (#399)
- (engine) stage a sink's indexes concurrently, and validate indexes concurrently
Fixed
- (engine) only an advance of the source is ingest progress; a failed apply drops its batch's buffer (#395 review 3)
- (source-postgres) read and write every number exactly
- (source-postgres) only a declared decimal changes what the sink receives
- (source-postgres) [breaking] write every numeric digit as a JSON number
- (source) a camelCase format name reads the ISO form; the date table follows the T form (#304)
- (sink) [breaking] write a naive timestamp as ISO 8601 with a T
Other
- a restart in the middle of a rebuild says rebuilding, vouching for nothing
- flusso.sink.refreshes{sink}: refresh requests a sink sent; dev/bench reports them per phase
- the sink keeps a backfill checkpoint: Sink::seeding, record_walked, forget_walked; OpenSearch keeps it in the meta doc
- caught up per lane: a caught-up write covers its quiet lanes' indexes, and a lane going quiet makes its own writes searchable (review 4, 5)
- a lane's SnapshotComplete seeds its index as a flight holding a write slot
- tests: ready items merge into one write (red); Sink::write_parts and write_limit
- (engine) an ordered sink gets its items in lane order, one at a time (#408 review 2, finding 4)
- (sink) [breaking] Sink::write replaces apply + flush (#386 stage 2)
- only a declared decimal may change what the sink receives
- a decimal keeps every digit; a non-finite number fails its document
- (query) a decimal past Decimal's range is written as a string (#306, #327)
- a naive timestamp must reach OpenSearch as ISO 8601
- (query-derive) null map values, date and constant kinds, named non-string types (#294)
- (query-derive) typed-map shapes the review found still compile and fail (#294)
flusso-sink-stdout-v0.18.0
Added
- (engine) an ordered sink gets lane order, one write at a time; stdout is one (#408 review 2, finding 4)
Fixed
- (source-postgres) [breaking] write every numeric digit as a JSON number
Other
- rustfmt
- Position carries its seq and the source's stamps (touches all constructors)
- (sink) [breaking] Sink::write replaces apply + flush (#386 stage 2)
flusso-sink-opensearch-v0.18.0
Added
- (engine) [breaking] stage timings account for each task's whole time (#399 review)
- per-stage timings in /metrics and the bench's attribution table (#399)
- (engine) stage a sink's indexes concurrently, and validate indexes concurrently
- (sink-opensearch) [breaking] warn when ca_file is set with tls_verify off; say when the file is read
- (sink-opensearch) trust a private CA with ca_file; never retry a refused certificate
Fixed
- (sink-opensearch) caught-up writes refresh one at a time (#408 review 1, finding 1)
- (engine) only an advance of the source is ingest progress; a failed apply drops its batch's buffer (#395 review 3)
- (daemon) reset an engine's restart backoff after a run that made progress (#395 review)
- (sinks) resend items OpenSearch refuses as busy, and release a lane delivery by its ticket (#395)
- (source-postgres) [breaking] write every numeric digit as a JSON number
- (engine,sink-opensearch) a rejected document's id is shown shortened; an empty-id delete is dropped
- (sink-opensearch) reject an id OpenSearch refuses instead of failing the bulk
Other
- a restart in the middle of a rebuild says rebuilding, vouching for nothing
- a first start goes on beside a plain index named like the alias; a generation without its meta doc is reused
- tests: a plain index named like the convenience alias, a first generation without its meta doc (red)
- flusso.sink.refreshes{sink}: refresh requests a sink sent; dev/bench reports them per phase
- the sink keeps a backfill checkpoint: Sink::seeding, record_walked, forget_walked; OpenSearch keeps it in the meta doc
- caught up per lane: a caught-up write covers its quiet lanes' indexes, and a lane going quiet makes its own writes searchable (review 4, 5)
- crash table for the new mark_seeded order; a first seed and each first-round request failing alone
- measured before and after; seeding indexes at once designed and pinned on the sink
- the sink engine merges ready items into one write
- rustfmt
- rustfmt mark_seeded
- cheaper seeding and generation creation (#410, #403)
- e2e for seeding and generation creation, request by request (#354, #403, #410)
- Position carries its seq and the source's stamps (touches all constructors)
- (sink-opensearch) each doc comment on its own item; every chunk of a caught-up write refreshes (#408 review 4, findings 1-2)
- (sink-opensearch) every chunk of a caught-up write refreshes, a multi-shard index is refreshed (#408 review 3, review 4 finding 2)
- (sink-opensearch) a caught-up write skips the refresh its own bulk already did (#408 review 2, finding 5)
- (sink-opensearch) a caught-up write over the one single-shard index left unrefreshed sends no extra refresh (#408 review 2, finding 5)
- (sink-opensearch) overlapping caught-up writes wait for an earlier refresh (#408 review 1, finding 1)
- (sink-opensearch) an empty write that isn't caught up reports Immediate (#386 stage 2)
- concurrent writes keep per-document order, the watermark and live-first (#386 stage 2)
- (sink) [breaking] Sink::write replaces apply + flush (#386 stage 2)
- a replayed change or backfill page is no progress, and a failed apply leaves nothing buffered (#395 review 3)
- an item refused as busy is resent, and a release redelivers only its ticket (#395)
- (sink-opensearch) write a bulk action line without a value tree
- (sink-opensearch) an id OpenSearch refuses fails the whole bulk request
flusso-script-v0.18.0
Added
- (script) [breaking] a real date type in script fields (#364)
Fixed
- (script,source-postgres) a declared float in --input is a float; jsonb zero padding is not digits
- (source-postgres) a float reads as Postgres casts to float4; exact reads pinned in flusso-query
- (source-postgres) an untyped number is a decimal to a script at every magnitude
- (source-postgres) read and write every number exactly
- (script) a thrown date is named as scripts name it
- (script) named zones reach every year chrono does; sort keeps Rhai's one-element rule (#364)
- (script) [breaking] no silent date comparisons; months as Postgres age(); zones from jiff (#364)
- (script) the nearest decimal past 28 digits; a zero decimal result has no sign
- (script) a declared decimal reaches a script as main read it, unless held exactly
- (script) a decimal script result is written with every digit
- (source-postgres) only a declared decimal changes what the sink receives
- (source-postgres) [breaking] write every numeric digit as a JSON number
Other
- the source marks a walk's point and resumes after it in its own key order (walk_mark, resume_after, compare_keys)
- a walk ends at the greatest key it found, read through the key index's own operators
- a backfill pages each index's root keys on the builder pool
- (script) calendar edges in named zones, sort parity with Rhai, the upgrade page read as written (#364)
- (script) refuse cross-kind date comparisons through Dynamic fallbacks (#364)
- (script) pin the review's silent answers, months as age(), and the upgrade replacements (#364)
- (script) point the schema skill at dates in a script; fix a private doc link (#364)
- (script) typed dates in scripts, failing first (#364)
- a script's decimal holds about 28 significant digits, not 28 fractional ones
- (script) a non-decimal value pays nothing for the decimal path
- (script) no precision cliff past 28 digits; a zero decimal result has no sign
- (script) a declared decimal reaches a script as main gave it
- (script) a decimal script result keeps its digits; a tiny decimal is a decimal
- only a declared decimal may change what the sink receives
flusso-query-v0.18.0
Added
- (query) [breaking] read-time fields: a root's context, rule fields and computed fields
- (query) [breaking] a fenced read rides out a restart of flusso
- (query-derive) read an array column as a Vec or an array newtype (#276)
- (query) Client::ca_file and ca_pem add a private CA to the pinned roots
Fixed
- (query) a rule reads a value the index coerced as the index stored it
- (query) no value, the value order and partial dates decide alike on both sides
- (query) jitter never drops a retry that fits; a spinning pacing test fails
- (query) pace, bound and classify fence retries (#377 review)
- (source-postgres) a float reads as Postgres casts to float4; exact reads pinned in flusso-query
- (source-postgres) read and write every number exactly
- (query-derive) an Arc/Rc field without
rcnames the feature (#306) - (query-derive) string alternatives, Decimal by its kinds, a true message (#306)
- (query-derive) a fragment accepts what a root accepts (#306)
- (query-derive) read null map values, refuse exhaustive enums without variants, name the tagged-newtype refusals
- (query-derive) check typed maps against what they read; pin the boundary
- (query) GeoPoint reads a compact format's fields as written (#307)
- (query) GeoPoint reads every form a geo field is written in
- (query-derive) [breaking] refuse a flattened non-object and a map type that reads no object
- (query-derive) [breaking] a newtype reads an array only as one of the two array kinds
- (query-derive) say why an array newtype does not fit a single value
- (query-derive) an array newtype of values holds leaf values only
- (query-derive) a value type is no sub-document at a nested or object field
- (query-derive) refuse an array of arrays at any depth of newtypes
- (query-derive) an array newtype needs a sub-document element
- (query-derive) [breaking] arrays of maps, newtypes over a Vec, hand-written values in fragments
- (query-derive) [breaking] root and fragment agree on maps; reject a rename serde does not see
- (query-derive) match a fragment's opaque to a root's, and close the review gaps
- (query-derive) [breaking] reject the remaining unread attributes
Other
- flusso-query: UnavailableCause::Backfilling
- docs: a slow index holds up only itself; a stop-refused document or a missing index still stops the sink (review 3)
- a configured lane holds the backstop, not the backstop times the sinks (review 11); search.rs doc order (review 9)
- docs: /wait?index, /status lanes, flusso.lane.outstanding, the client's per-index fence, upgrade page, CLAUDE.md, CONTEXT.md, stage notes as built
- query: a fenced read waits on the indexes it reads; the cache is per level and index
- Merge remote-tracking branch 'origin/main' into feat/read-time-fields
- (query) read-time fields
- (query) keep the test stub's TLS key out of the package
- (query-derive) no assert on a field type named Arc or Rc (#306)
- Merge pull request #300 from alias2k/fix/derive-map-boundary-294
- (query) a GeoPoint written by a compact format reads back as written (#307)
- (query) GeoPoint reads every form a geo field is written in
Changed
- (query) a fenced read waits on the indexes it reads (
/wait?index=…), once per level and index;Client::wait_onwaits on named indexes;MsearchBundle::indexes(default none: every index)
flusso-query-derive-v0.18.0
Added
- (query) [breaking] read-time fields: a root's context, rule fields and computed fields
- (query-derive) read an array column as a Vec or an array newtype (#276)
Fixed
- (query-derive) an Arc/Rc field without
rcnames the feature (#306) - (query-derive) string alternatives, Decimal by its kinds, a true message (#306)
- (query-derive) a fragment accepts what a root accepts (#306)
- (source) a camelCase format name reads the ISO form; the date table follows the T form (#304)
- (query-derive) refuse by name only what can't be a string type of your own
- (query-derive) say only what the derive promises; make the boundary test cheap
- (query-derive) read null map values, refuse exhaustive enums without variants, name the tagged-newtype refusals
- (query-derive) close the typed-map holes the review found; pin root/fragment agreement
- (query-derive) check typed maps against what they read; pin the boundary
- (kernel) store a constant list mixing kinds unindexed, on a new index
- (query-derive) two refusals suggest a fix that works
- (query-derive) [breaking] refuse a flattened non-object and a map type that reads no object
- (query-derive) [breaking] a newtype reads an array only as one of the two array kinds
- (query-derive) a fragment takes what a root takes at a json array
- (query-derive) a newtype over an array newtype reads the array too
- (query-derive) say why an array newtype does not fit a single value
- (query-derive) an array newtype of values holds leaf values only
- review findings on array columns (#276)
- (query-derive) a newtype over a primitive is a value; opaque in a fragment keeps the kind check at a level
- (query-derive) a value type is no sub-document at a nested or object field
- (query-derive) refuse an array of arrays at any depth of newtypes
- (query-derive) an array newtype needs a sub-document element
- (query-derive) [breaking] arrays of maps, newtypes over a Vec, hand-written values in fragments
- (query-derive) a field serde skips may still name its level with flusso(rename)
- (query-derive) [breaking] root and fragment agree on maps; reject a rename serde does not see
- (query-derive) a transparent named-field struct is a value newtype too
- (query-derive) match a fragment's opaque to a root's, and close the review gaps
- (query-derive) [breaking] reject the remaining unread attributes
- (query-derive) leave a flattened catch-all unchecked in a fragment; reject a rename serde contradicts
- (query-derive) [breaking] reject opaque, rename and scope_mod where they do nothing
- (query-derive) reject a second scope for one level; the flatten error names the way out
- (query-derive) [breaking] a scope attribute that renames nothing is a compile error
Other
- (query-derive) a rule handle for every kind, and each way a stored handle is renamed
- (query-derive) read-time codegen in its own module
- (query) read-time fields
- (query-derive) read-time fields through the derive, and every refusal
- (query-derive) a past-range decimal is a number, which f64 reads
- (query-derive) no assert on a field type named Arc or Rc (#306)
- (query) a decimal past Decimal's range is written as a string (#306, #327)
- (query-derive) pin string alternatives and foreign types at custom (#306)
- (query-derive) review round of #306
- (query-derive) state and pin every trusted rule of the derive (#306)
- (query-derive) a fragment accepts what a root accepts (#306)
- (query-derive) string-reading map keys and tagged newtypes the derive refuses by name (#294)
- (query-derive) null map values, date and constant kinds, named non-string types (#294)
- (query-derive) typed-map shapes the review found still compile and fail (#294)
- (query-derive) typed maps the derive does not check yet (#294)
- (query-derive) drop the Box case, whose rustc notes differ between hosts