Repository navigation
fix: collect long-text results one at a time, so a long line no longer overflows the stack - #2248
Merged
Merged
Conversation
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
… of ; or ( is refused A line of ; or ( places one refusal per character, and splitWords added a line's refusals to the text's list as the arguments of one call, which overflows the stack past the arguments a runtime takes: isDestructiveNonSqlQuery raised RangeError at 700,000 such characters under Bun 1.4.4 and at 130,000 under Node 24.22.1, so the web editor's gate and the provider's own guard failed on the line, and POST /api/db/statement/analyze on the production server answered it 500. lexLogicalLine collected its refusals the same way. Both now push them in a loop, and the lexer's output is unchanged. tests/unit/db/etcd/long-line-runtimes.test.ts drives a 1,000,000-character line of ; and one of ( through the gate and the provider's guard under Bun and in a Node child, and each is refused at its first character, as a line of one is. The lexer and route tests hold the same lines. Backlog entry D453 is removed.
…options key refuseOptions pushed a list's items onto its walk as the arguments of one call. The console admits a list of 262,144 numbers, and that walk overflowed the stack at 130,000 under Node 24.22.1, so the gate, the statement analysis and the provider, which all run it first, failed on a long vector query under Node, and POST /api/db/statement/analyze on the production server answered it 500. Bun 1.4.4 took the whole bound. The walk now pushes the items in a loop, in the same order. tests/unit/db/qdrant/request-runtimes.test.ts gains a vector of the console's 262,144 numbers, accepted with the same body on Bun and on Node.
cypherCompletionContext added each line's tokens to its list as the arguments of one call, so a line of one token per character overflowed the stack: at 700,000 under Bun 1.4.4 and at 130,000 under Node 24.22.1, the V8 of Chrome and Edge. The tokens are now pushed in a loop, and a line of 1,000,000 reads to the context its end is in.
statementScope added a level's sources and its CTE names to its lists as the arguments of one call, so a long enough FROM list or WITH list overflowed the stack in the SQL completion provider: at 700,000 under Bun 1.4.4 and at 130,000 under Node 24.22.1, the V8 of Chrome and Edge. Both are now pushed in a loop, and a list of 1,000,000 names every item.
Under Bun 1.4.4 the original code passed 600,000 and overflowed from 650,000; the comments named Node's 130,000 instead.
…Redis argument spread D456: a line of ; or ( grows the Node heap by several hundred bytes per character and aborts the process at 10,000,000 characters, under the body limit. D457: runCommand spreads a command's words into callBuffer and answers the overflow as a Redis error.
cevheri
force-pushed
the
fix/long-text-argument-spreads
branch
from
November 4, 2026 22:52
1bc285b to
f504b13
Compare
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes D453. A long user text could overflow the stack where a reader spread a text-sized array into
push(...): the etcd lexer, the Qdrant request reader, the Cypher completion tokens and the SQL scope reader. ThroughPOST /api/db/statement/analyzeunder Node 24, etcd;x 130,000 and a Qdrant vector of 130,000 numbers answered 500Maximum call stack size exceeded; they now answer 200, and etcd;x 1,000,000 answers in about 300 ms.Each fix replaces the spread with a loop; output and order are unchanged, and every new test failed before its fix (including a Node child process, since Bun's stack is deeper). The spread sites judged unreachable by a user text are listed in the backlog entry's resolution. Two related findings are filed: the etcd lexer's memory on one very long line, and a Redis argument spread.
Gate set green, coverage 100%.