11355 commits

This branch

This branch

All branches

Author	SHA1	Message	Date
Jussi Saurio	c7672b952b	Use Cow for Value::Blob to prevent copies in op_column	2025-12-06 12:08:27 +02:00
Jussi Saurio	2742278e5a	decouple functions from struct Value	2025-12-06 10:42:01 +02:00
Jussi Saurio	299ccdaee4	Merge 'sim: stop ignoring sql execution errors' from Jussi Saurio ... Reviewed-by: Pedro Muniz (@pedrocarlo) Closes #4106	2025-12-05 23:04:20 +02:00
Preston Thorpe	231282a5be	Merge 'Simulator Roadmap' from Alperen Keleş ... This PR is a working doc on a roadmap for the simulator. @pedrocarlo @LeMikaelF please take a look. Reviewed-by: Preston Thorpe <preston@turso.tech> Closes #3954	2025-12-05 13:27:57 -05:00
Preston Thorpe	c09c30746e	Merge 'guard subjournal access within single connection' from Nikita Sivukhin ... Right now turso can panic with various asserts if 2 or more write statements will be executed over single connection concurrently: ``` thread 'query_processing::test_write_path::api_misuse' panicked at core/storage/pager.rs:776:9: subjournal offset should be 0 ``` This PR adds explicit guard for subjournal access which will return `Busy` for the operation internally and lead to wait condition for the statement until subjournal ownership will be released and can be re- acquired again. Reviewed-by: Jussi Saurio <jussi.saurio@gmail.com> Closes #4110	2025-12-05 13:14:07 -05:00
Preston Thorpe	e7c7f232b4	Merge 'testing/fuzz: Add new fuzzer for joins' from Preston Thorpe ... needed for #4063 to merge, currently passing on main but just want to lower the already huge diff Reviewed-by: Jussi Saurio <jussi.saurio@gmail.com> Closes #4103	2025-12-05 13:13:44 -05:00
Nikita Sivukhin	659ef7c079	fix clippy	2025-12-05 21:39:35 +04:00
Nikita Sivukhin	487854e6d6	guard subjournal access in order to prevent concurrent operations over it within same connection	2025-12-05 21:25:13 +04:00
PThorpe92	c9a6827011	Extract out join fuzzer to an additional test on indexed columns	2025-12-05 12:03:23 -05:00
Preston Thorpe	97eb482885	Merge 'Fix two bugs with compound selects' from Jussi Saurio ... Closes #4108 1. Previously compound select would not start a transaction if the rightmost subselect had no table references, e.g. SELECT * FROM t UNION VALUES(1) 2. Previously the column names for the query were taken from the rightmost subselect - instead, they should be taken from the leftmost subselect. Reviewed-by: Preston Thorpe <preston@turso.tech> Closes #4109	2025-12-05 11:51:38 -05:00
Jussi Saurio	58a25f7f5f	Merge 'Turso sdk kit version' from Nikita Sivukhin ... This PR adds simple `turso_version()` function to the sdk-kit crate Closes #4099	2025-12-05 17:27:28 +02:00
Jussi Saurio	a90087bcf6	Enable compound_select_fuzz for mvcc because it works as a regression test for #4108	2025-12-05 17:19:05 +02:00
Jussi Saurio	e0791406b5	Fix two bugs with compound selects ... 1. Previously compound select would not start a transaction if the rightmost subselect had no table references, e.g. SELECT * FROM t UNION VALUES(1) 2. Previously the column names for the query were taken from the rightmost subselect - instead, they should be taken from the leftmost subselect.	2025-12-05 17:14:58 +02:00
Jussi Saurio	39325fdca9	Merge 'Subsec failed to format as YYYY-MM-DD HH:MM:SS.SSS' from ... Sqlite docs reference: https://sqlite.org/lang_datefunc.html The datetime() function returns the date and time formatted as YYYY-MM- DD HH:MM:SS or as YYYY-MM-DD HH:MM:SS.SSS if the [subsec modifier](https://sqlite.org/lang_datefunc.html#subsec) is used. ### Failed before changes: ``` sql INSERT INTO test_results VALUES ( 'Fixed DateTime Expansion', datetime('2024-01-01 12:00:00', 'subsec'), '2024-01-01 12:00:00.000', CASE WHEN datetime('2024-01-01 12:00:00', 'subsec') = '2024-01-01 12:00:00.000' THEN 'PASS' ELSE 'FAIL' END, 'Adds .000 to fixed time' ); ``` <img width="1164" height="94" alt="Screenshot 2025-12-04 130827" src="https://github.com/user- attachments/assets/d09dbe19-d329-4b88-a727-8f92e79e0a10" /> ``` sql INSERT INTO test_results VALUES ( 'Date Only Input', datetime('2024-01-01', 'subsec'), '2024-01-01 00:00:00.000', CASE WHEN datetime('2024-01-01', 'subsec') = '2024-01-01 00:00:00.000' THEN 'PASS' ELSE 'FAIL' END, 'Expands date to midnight.000' ); INSERT INTO test_results VALUES ( 'ISO "T" Separator', datetime('2024-01-01T15:30:00', 'subsec'), '2024-01-01 15:30:00.000', CASE WHEN datetime('2024-01-01T15:30:00', 'subsec') = '2024-01-01 15:30:00.000' THEN 'PASS' ELSE 'FAIL' END, 'Replaces T with space' ); INSERT INTO test_results VALUES ( 'Chain: Subsec then Add', datetime('2024-01-01 12:00:00', 'subsec', '+1 hour'), '2024-01-01 13:00:00.000', CASE WHEN datetime('2024-01-01 12:00:00', 'subsec', '+1 hour') = '2024-01-01 13:00:00.000' THEN 'PASS' ELSE 'FAIL' END, 'Flag persists after +1 hour' ); INSERT INTO test_results VALUES ( 'Chain: Add then Subsec', datetime('2024-01-01 12:00:00', '+1 hour', 'subsec'), '2024-01-01 13:00:00.000', CASE WHEN datetime('2024-01-01 12:00:00', '+1 hour', 'subsec') = '2024-01-01 13:00:00.000' THEN 'PASS' ELSE 'FAIL' END, 'Standard order works' ); INSERT INTO test_results VALUES ( 'Case Insensitivity', datetime('2024-01-01 12:00:00', 'SuBsEc'), '2024-01-01 12:00:00.000', CASE WHEN datetime('2024-01-01 12:00:00', 'SuBsEc') = '2024-01-01 12:00:00.000' THEN 'PASS' ELSE 'FAIL' END, 'SuBsEc works' ); ``` <img width="1206" height="340" alt="Screenshot 2025-12-04 131114" src="https://github.com/user- attachments/assets/72fd7fe2-c991-4a92-b3d7-2f4516404fa7" /> Reviewed-by: Preston Thorpe <preston@turso.tech> Closes #4097	2025-12-05 15:31:56 +02:00
Jussi Saurio	ca605b368b	Merge 'Fix IN operator translation logic' from Nikita Sivukhin ... This PR replaces incorrect usage of `program.resolve_label(...)` to the correct method `program.preassign_label_to_next_insn(...)` for IN operator translation code. Following query translated with bug before the fix: ```sql turso> CREATE TABLE t (id INTEGER PRIMARY KEY, name TEXT); turso> EXPLAIN SELECT COUNT(*) FROM t WHERE name in ('alice', 'bob'); ``` Before `emit_constant_insns` optimization query plan was correct ```sql turso> EXPLAIN SELECT COUNT(*) FROM t WHERE name in ('alice', 'bob'); addr opcode p1 p2 p3 p4 p5 comment ---- ----------------- ---- ---- ---- ------------- -- ------- 0 Init 0 16 0 0 Start at 16 1 Null 0 2 0 0 r[2]=NULL 2 OpenRead 0 2 0 0 table=t, root=2, iDb=0 3 Rewind 0 12 0 0 Rewind table t 4 Column 0 1 3 0 r[3]=t.name 5 String8 0 4 0 alice 0 r[4]='alice' 6 Eq 3 4 9 Binary 0 if r[3]==r[4] goto 9 7 String8 0 5 0 bob 0 r[5]='bob' 8 Ne 3 5 11 Binary 0 if r[3]!=r[5] goto 11 9 Integer 1 6 0 0 r[6]=1 10 AggStep 0 6 2 count 0 accum=r[2] step(r[6]) 11 Next 0 4 0 0 12 AggFinal 0 2 0 count 0 accum=r[2] 13 Copy 2 1 0 0 r[1]=r[2] 14 ResultRow 1 1 0 0 output=r[1] 15 Halt 0 0 0 0 16 Transaction 0 1 1 0 iDb=0 tx_mode=Read 17 Goto 0 1 0 0 ``` But the problem is that after `emit_constant_insns`, `Eq` jump target was rewritten as it was binded to the 9th op code (Integer 1) instead of the next op code after the IN translated block (in the final plan, note jump to the 16 address for Eq instruction) ```sql turso> EXPLAIN SELECT COUNT(*) FROM t WHERE name in ('alice', 'bob'); addr opcode p1 p2 p3 p4 p5 comment ---- ----------------- ---- ---- ---- ------------- -- ------- 0 Init 0 13 0 0 Start at 13 1 Null 0 2 0 0 r[2]=NULL 2 OpenRead 0 2 0 0 table=t, root=2, iDb=0 3 Rewind 0 9 0 0 Rewind table t 4 Column 0 1 3 0 r[3]=t.name 5 Eq 3 4 16 Binary 0 if r[3]==r[4] goto 16 6 Ne 3 5 8 Binary 0 if r[3]!=r[5] goto 8 7 AggStep 0 6 2 count 0 accum=r[2] step(r[6]) 8 Next 0 4 0 0 9 AggFinal 0 2 0 count 0 accum=r[2] 10 Copy 2 1 0 0 r[1]=r[2] 11 ResultRow 1 1 0 0 output=r[1] 12 Halt 0 0 0 0 13 Transaction 0 1 1 0 iDb=0 tx_mode=Read 14 String8 0 4 0 alice 0 r[4]='alice' 15 String8 0 5 0 bob 0 r[5]='bob' 16 Integer 1 6 0 0 r[6]=1 17 Goto 0 1 0 0 ``` Reviewed-by: Jussi Saurio <jussi.saurio@gmail.com> Closes #4105	2025-12-05 15:31:32 +02:00
Jussi Saurio	74296e52bb	Merge 'Automatically Propagate Encryption options' from Pedro Muniz ... On database open, we store the Encryption Options and pass them onwards to the Connection, Pager and Wal. We also have slight gain in ergonomics, as we don't have set the Pragma's for the `cipher` and `hexkey` on each new `Connection`. I needed this logic, because I will need to initialize a Default Header for empty DBs and encryption opts not being automatically propagated was hindering me for this. **Ai Disclosure** Claude helped me debug and find out issues in my implementation cc @avinassh Reviewed-by: Avinash Sajjanshetty (@avinassh) Closes #4100	2025-12-05 15:31:17 +02:00
Nikita Sivukhin	d5f58de801	fix clippy	2025-12-05 15:29:17 +04:00
Jussi Saurio	77ad9d87c5	sim: stop ignoring sql execution errors	2025-12-05 13:14:39 +02:00
Nikita Sivukhin	17695438ac	fix IN operator translation bug ... - following query translated with bug before the fix: turso> CREATE TABLE t (id INTEGER PRIMARY KEY, name TEXT); turso> EXPLAIN SELECT COUNT(*) FROM t WHERE name in ('alice', 'bob'); - before emit_constant_insns optimization query plan was correct turso> EXPLAIN SELECT COUNT(*) FROM t WHERE name in ('alice', 'bob'); addr opcode p1 p2 p3 p4 p5 comment ---- ----------------- ---- ---- ---- ------------- -- ------- 0 Init 0 16 0 0 Start at 16 1 Null 0 2 0 0 r[2]=NULL 2 OpenRead 0 2 0 0 table=t, root=2, iDb=0 3 Rewind 0 12 0 0 Rewind table t 4 Column 0 1 3 0 r[3]=t.name 5 String8 0 4 0 alice 0 r[4]='alice' 6 Eq 3 4 9 Binary 0 if r[3]==r[4] goto 9 7 String8 0 5 0 bob 0 r[5]='bob' 8 Ne 3 5 11 Binary 0 if r[3]!=r[5] goto 11 9 Integer 1 6 0 0 r[6]=1 10 AggStep 0 6 2 count 0 accum=r[2] step(r[6]) 11 Next 0 4 0 0 12 AggFinal 0 2 0 count 0 accum=r[2] 13 Copy 2 1 0 0 r[1]=r[2] 14 ResultRow 1 1 0 0 output=r[1] 15 Halt 0 0 0 0 16 Transaction 0 1 1 0 iDb=0 tx_mode=Read 17 Goto 0 1 0 0 - but the problem is that after emit_constant_insns Eq jump target was rewritten as it was binded to the 9th op code (Integer 1) instead of the next op code after the IN translated block (in the final plan, note jump to the 16 address for Eq instruction) turso> EXPLAIN SELECT COUNT(*) FROM t WHERE name in ('alice', 'bob'); addr opcode p1 p2 p3 p4 p5 comment ---- ----------------- ---- ---- ---- ------------- -- ------- 0 Init 0 13 0 0 Start at 13 1 Null 0 2 0 0 r[2]=NULL 2 OpenRead 0 2 0 0 table=t, root=2, iDb=0 3 Rewind 0 9 0 0 Rewind table t 4 Column 0 1 3 0 r[3]=t.name 5 Eq 3 4 16 Binary 0 if r[3]==r[4] goto 16 6 Ne 3 5 8 Binary 0 if r[3]!=r[5] goto 8 7 AggStep 0 6 2 count 0 accum=r[2] step(r[6]) 8 Next 0 4 0 0 9 AggFinal 0 2 0 count 0 accum=r[2] 10 Copy 2 1 0 0 r[1]=r[2] 11 ResultRow 1 1 0 0 output=r[1] 12 Halt 0 0 0 0 13 Transaction 0 1 1 0 iDb=0 tx_mode=Read 14 String8 0 4 0 alice 0 r[4]='alice' 15 String8 0 5 0 bob 0 r[5]='bob' 16 Integer 1 6 0 0 r[6]=1 17 Goto 0 1 0 0	2025-12-05 14:52:09 +04:00
Nikita Sivukhin	e839eb499b	make fuzzer to generate SELECT COUNT(*) OR SELECT * statements ... - this is important for IN operation translation bug because in case of COUNT(*) there is constant assignment instruction right after last instruction translated from IN condition	2025-12-05 14:48:59 +04:00
Jussi Saurio	6da0895813	Merge 'fix/mvcc: seek() must seek from both mv store and btree' from Jussi Saurio ... Closes #4093 for example, upon opening an existing database, all the rows are in the btree, so if we seek only from MV store, we won't find anything. ergo: we must look from both the mv store and the btree. if we are iterating forwards, the smallest of the two results is where we land, and vice versa for backwards iteration. --- this PR was almost entirely written using Claude Code + Opus 4.5, and here is Claude's gangsta rap about the changes: > Can you write a gangsta rap that summarizes everything you changed in this session? I.e. what was fixed and how Here's the audio version generated using Suno: https://suno.com/song/2209deaf-4c59-43c5-b0e9-d807b9805dc0 ⏺ "Dual Seek Chronicles" (verse 1) Yo, the bug was real, seek was half-baked, Only checked MVCC, btree forsaked, User said "look, when both stores got data, Seek gotta check both—fix it now, not later." (pre-chorus) See the cursor got two sources of truth, Btree on disk, MVCC in the booth, After checkpoint, data splits in two, Seek was blind to half—that just won't do. (chorus) Now we seek in MVCC, seek in btree, Pick the winner based on which direction we be, Forwards? Smallest key wins the race, Backwards? Biggest key takes first place. Dual iteration, that's the game we play, seek_btree_and_set_peek saves the day! (verse 2) But wait—SeekResult got three states to check, Found means we good, NotFound means we wreck, TryAdvance though? Cursor's positioned but stuck, Gotta call next() or prev() for luck. Added helper methods, kept the code tight, advance_btree_forward_from_current done right, But then we noticed duplication in the mix, Refactored with a bool—that's the slickest fix. (bridge) _advance_btree_forward(initialize: bool), One function to rule them, that's the protocol, true means rewind if uninitialized state, false means we seeked—just iterate. (verse 3) Dropped six tests in test_transactions.rs, Checkpoint, insert, seek—put it to the test, Interleaved rows, odd in btree, even in MV, Updates and deletes, shadow what you see. Range operations: GT, LT, GE, LE, All forty-three tests pass, we're bug-free, From Uninitialized to Row or Exhausted state, Turso's dual cursor finally running straight. (outro) That's the story of the dual-seek fix, MVCC and btree no longer in conflicts, Ship it to main, let the users feast, Cursor seeking both stores—east to west, west to east. 🎤✨ EDIT: After merge of #4095 , I refactored this PR to use state machines instead of blocking IO as well. Initial implementation was all Claude Code + Opus 4.5 there too, but the state machine code was FAR too complex so I manually cleaned up and simplified it. Closes #4094	2025-12-05 12:07:49 +02:00
Jussi Saurio	eb782ce2d4	fix/mvcc: seek() must seek from both mv store and btree ... for example, upon opening an existing database, all the rows are in the btree, so if we seek only from MV store, we won't find anything. ergo: we must look from both the mv store and the btree. if we are iterating forwards, the smallest of the two results is where we land, and vice versa for backwards iteration. initially this implementation used blocking IO but was refactored to use state machines after the rest of the Cursor methods in the MVCC cursor module were refactored to do that too. --- this PR was initially almost entirely written using Claude Code + Opus 4.5, but heavily manually cleaned up as the AI made the state machine refactor far too complicated.	2025-12-05 11:53:16 +02:00
Jussi Saurio	8f57c60c26	Merge 'core/mvcc: state machines for prev, next, exists, rewind, last' from Pere Diaz Bou ... Basically bring back state machines for: `next`, `prev`, `last`, `rewind`, `get_next_rowid`. Reviewed-by: Jussi Saurio <jussi.saurio@gmail.com> Closes #4095	2025-12-05 10:08:00 +02:00
pedrocarlo	ee73bab743	get correct reserved bytes if Cipher is not None	2025-12-05 02:04:06 -03:00
pedrocarlo	a311c966a2	set encryption context for page and wal in init_pager	2025-12-05 02:04:06 -03:00
pedrocarlo	889322f6b5	do not call pragmas related to encryption on connect or open	2025-12-05 02:04:06 -03:00
pedrocarlo	0118a65169	pass encryption opts from the database to the connection on connect	2025-12-05 02:04:06 -03:00
pedrocarlo	85b212056d	separate init function for connect	2025-12-05 02:04:06 -03:00
pedrocarlo	1a43de35ce	add encryption key and cipher to Database struct	2025-12-05 02:04:06 -03:00
pedrocarlo	faca85de2f	pass pager to _connect and share initial coon for boostrapping mvcc	2025-12-05 02:04:05 -03:00
Preston Thorpe	5b3262ba18	Merge 'add lib-release profile' from Nikita Sivukhin ... This profile will provide another trade-off between performance and library final size Reviewed-by: Preston Thorpe <preston@turso.tech> Closes #4096	2025-12-04 18:40:25 -05:00
Preston Thorpe	227c8ad5d0	Merge 'planner/vdbe: implement Hash Joins as an alternative to Ephemeral Indexes ' from Preston Thorpe ... This PR adds a hash table data structure that partitions and spills to disk when a memory limit is exceeded, which is used (a bit conservatively) with the following opcodes for the VDBE: `HashBuild`: feed a build-side row into the hash table. `HashBuildFinalize`: finalize the build, potentially spilling partitions. `HashProbe`: hash and probe a key, yielding the build-side rowid(s). `HashNext`: iterate through additional matches in the same bucket. `HashClose`: free the hash table. There is pretty extensive doc comments for both the hash table implementation and for the planner/access method selection. The basic idea (for now) is that one ephemeral index is replaced with a hash table and we get So far it's showing significant perf increase: ### Query: ```sql select users.id, products.id, o.id from users join products on products.name = users.first_name join order_items o on o.unit_price = products.price; ``` **Before**: `total: 20.918470388 s` **After**: `total: 289 ms` (this is with `main` building an ephemeral index for one but doing a full scan for two outermost tables) #### Bytecode ```sql addr opcode p1 p2 p3 p4 p5 comment ---- ----------------- ---- ---- ---- ------------- -- ------- 0 Init 0 47 0 0 Start at 47 1 OpenRead 0 2 0 0 table=users, root=2, iDb=0 2 OpenRead 1 3 0 0 table=products, root=3, iDb=0 3 OpenRead 2 72 0 0 table=order_items, root=72, iDb=0 4 Rewind 1 45 0 0 Rewind table products 5 Once 14 0 0 0 goto 14 6 OpenRead 4 2 0 0 =users, root=2, iDb=0 7 Rewind 4 13 0 0 Rewind users 8 Column 4 1 4 0 r[4]=users.first_name 9 RowId 4 5 0 0 r[5]=users.rowid 10 Column 4 1 6 0 r[6]=users.first_name 11 HashBuild 4 4 1 r=[1] budget=67108864 payload=r[5]..r[6] 0 12 Next 4 8 0 0 13 HashBuildFinalize 1 0 0 0 14 Column 1 1 7 0 r[7]=products.name 15 HashProbe 1 7 1 r[10]=44 payload=r[8]..r[9] 0 16 Once 25 0 0 0 goto 25 17 OpenAutoindex 3 0 0 0 cursor=3 18 Rewind 2 19 0 0 Rewind table order_items 19 Column 2 4 11 0 r[11]=order_items.unit_price 20 RowId 2 12 0 0 r[12]=order_items.rowid 21 RowId 2 13 0 0 r[13]=order_items.rowid 22 MakeRecord 11 3 14 0 r[14]=mkrec(r[11..13]); for ephemeral_order_items_t3 23 IdxInsert 3 14 11 0 key=r[14] 24 Next 2 19 0 0 25 Column 1 2 15 0 r[15]=products.price 26 RealAffinity 15 0 0 0 27 IsNull 15 42 0 0 if (r[15]==NULL) goto 42 28 Affinity 15 1 0 0 r[15..16] = C 29 SeekGE 3 42 15 0 key=[15..15] 30 IdxGT 3 42 15 0 key=[15..15] 31 DeferredSeek 3 2 0 0 32 Column 3 0 17 0 r[17]=ephemeral_order_items_t3.unit_price 33 RealAffinity 17 0 0 0 34 Column 1 2 18 0 r[18]=products.price 35 RealAffinity 18 0 0 0 36 Ne 17 18 41 Binary 0 if r[17]!=r[18] goto 41 37 Copy 8 1 0 0 r[1]=r[8] 38 RowId 1 2 0 0 r[2]=products.rowid 39 IdxRowId 3 3 0 0 r[3]=cursor 3 for index ephemeral_order_items_t3.rowid 40 ResultRow 1 3 0 0 output=r[1..3] 41 Next 3 30 0 0 42 HashNext 1 10 44 payload=r[8]..r[9] 0 43 Goto 0 16 0 0 44 Next 1 14 0 0 45 HashClose 1 0 0 0 46 Halt 0 0 0 0 47 Transaction 0 1 28 0 iDb=0 tx_mode=Read 48 Goto 0 1 0 0 ``` Comparison with `sqlite3`: **query**: ```sql select users.id, products.id from users join products on products.name = users.first_name; ``` 15k rows in each table, no index on either column sqlite: `22ms` tursodb: `7ms` I ran the simulator for 5000 tests each for a run of 200 and all of them passed.. keeping in mind that the default limit at which a table will spill to disk for debug build is only 32kb so it should be frequent. Reviewed-by: Jussi Saurio <jussi.saurio@gmail.com> Closes #4063	2025-12-04 18:38:06 -05:00
PThorpe92	170ed26732	Fix NULL handling of hash join comparison and bug in cursor override logic	2025-12-04 17:21:46 -05:00
PThorpe92	16a1940d3a	Reduce iteration count in join fuzzer to 2000	2025-12-04 16:23:46 -05:00
PThorpe92	bf6038f0ba	Add join fuzzer for non-indexed columns	2025-12-04 16:19:19 -05:00
PThorpe92	c0d35dbaba	Add some additional join tests with nulls for new bloom filter behavior	2025-12-04 16:09:49 -05:00
PThorpe92	d0f15d1537	Fix bloom filter impl to skip NULL values	2025-12-04 16:09:49 -05:00
PThorpe92	4225bec5e9	Apply some review suggestions	2025-12-04 16:09:49 -05:00
PThorpe92	f3affbba2b	Fix condition evaluation that references hash build table now that its no longer in join order	2025-12-04 16:09:49 -05:00
PThorpe92	e8ef5dbf06	Cache result column values in the hashtable to prevent additional SeekRowID	2025-12-04 16:09:48 -05:00
PThorpe92	ad7d34bb67	Make hash joins follow pattern of ephemeral indexes instead of hoisting special logic	2025-12-04 16:09:48 -05:00
PThorpe92	eebb4950c6	Apply some review suggestions/cleanups	2025-12-04 16:09:48 -05:00
PThorpe92	866b153a7a	Fix clippy warnings	2025-12-04 16:09:47 -05:00
PThorpe92	746fe82159	Add/update some comments in the query planner	2025-12-04 16:09:47 -05:00
PThorpe92	06cbe8c063	Add more py tests for hash joins	2025-12-04 16:09:47 -05:00
PThorpe92	bf61feb6bc	Fix hash join planning/emission for > 2 way joins	2025-12-04 16:09:47 -05:00
PThorpe92	fd93544af3	Remove unneeded label resolution in close_loop method for hash join build tbl	2025-12-04 16:09:46 -05:00
PThorpe92	9b274d9243	Prevent using hash join for any query with an index on a join key	2025-12-04 16:09:46 -05:00
PThorpe92	bc5406be04	Add memory py tests for hash spilling, fix/finish spilling impl in hashtable	2025-12-04 16:09:46 -05:00
PThorpe92	bea9648a98	Fix hash join spilling for async IO by keeping buffer alive in callback	2025-12-04 16:09:45 -05:00