Repository navigation
mcp: make the ioConn frame cap independent of read boundaries - #1510
Merged
guglielmo-san merged 2 commits intoNov 6, 2026
Merged
Conversation
frameLimitReader reset its count after each message, but json.Decoder reads ahead: when it finishes a message it may already hold bytes of the next one, and those bytes were counted toward the finished message. The next message could then exceed MaxLineLength by as many bytes as had been read ahead, so whether it was rejected depended on how the reader split the stream. The reader now counts the bytes it has read, and each frame's budget starts at the decoder's InputOffset, so whether a message fits depends only on the message. That is what the MaxLineLength documentation already describes, so the documentation is unchanged. Line terminators and blank lines before a message count toward it, since they follow the previous message. Without the change, four cases of the new table test fail, and the fuzz test finds a counterexample within a second. Fixes modelcontextprotocol#1509 AI assistance: prepared with Claude; I reviewed the change and can explain it.
guglielmo-san
approved these changes
Nov 6, 2026
Tangerg
added a commit
to Tangerg/go-sdk
that referenced
this pull request
Nov 6, 2026
Sync 6 upstream commits through df0f295: the new ParseURLElicitationRequiredError (modelcontextprotocol#1495), the tools/call retry on HeaderMismatch (modelcontextprotocol#1502), notifications without Mcp-Method (modelcontextprotocol#1504), the cancelable ioConn write lock (modelcontextprotocol#1499), invalid_token in 401 challenges (modelcontextprotocol#1507) and the ioConn frame cap that no longer depends on read boundaries (modelcontextprotocol#1510). modelcontextprotocol#1499 and modelcontextprotocol#1510 are the upstream form of the fork's transport fixes, so ioConn now matches upstream except that canceling a write already in progress still closes the connection, because the peer may hold part of the frame. Upstream's tests replace the fork's duplicates of them; the fork keeps its test for the in-progress cancellation. New pkg.go.dev links in the docs use the fork's module path.
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.
frameLimitReader reset its count after each message, but json.Decoder
reads ahead: when it finishes a message it may already hold bytes of the
next one, and those bytes were counted toward the finished message. The
next message could then exceed MaxLineLength by as many bytes as had
been read ahead, so whether it was rejected depended on how the reader
split the stream.
The reader now counts the bytes it has read, and each frame's budget
starts at the decoder's InputOffset, so whether a message fits depends
only on the message. That is what the MaxLineLength documentation
already describes, so the documentation is unchanged. Line terminators
and blank lines before a message count toward it, since they follow the
previous message.
Without the change, four cases of the new table test fail, and the fuzz
test finds a counterexample within a second.
Fixes #1509
AI assistance: prepared with Claude; I reviewed the change and can
explain it.