[fix] I/P: Fix connection failure during some remote builds#201
Open
hanno-becker wants to merge 9 commits intomainfrom
Open
[fix] I/P: Fix connection failure during some remote builds#201hanno-becker wants to merge 9 commits intomainfrom
hanno-becker wants to merge 9 commits intomainfrom
Conversation
Symptom: Sessions with multiple `theories` blocks in ROOT failed when built remotely via I/P, with repeated "Connection refused: failed to open socket 127.0.0.1:<port>" errors. The build completed all theory loading but exited with rc=1. Issue: When proxying a build to a remote machine, the proxy rewrites bash_process_address in PIDE messages so that the remote ML process connects to a remote Bash.Server instead of the local one. The build_session message carries a separate Options object per theory group, each containing bash_process_address. The rewrite only replaced the first occurrence (count=1), leaving subsequent groups with the unrewritten local address. Fix: Remove the count limit from replace() for both bash_process_address and bash_process_password, so all theory groups are rewritten. Signed-off-by: Hanno Becker <beckphan@amazon.co.uk>
Source file verification was skipped entirely when the variable was missing. Now defaults to the resolved --dir (or pwd), shows a warning, and proceeds with source status reporting. Signed-off-by: Hanno Becker <beckphan@amazon.co.uk>
This allows an agent with I/R MCP connection to spawn multiple subagents working concurrently in separate REPLs. Signed-off-by: Hanno Becker <beckphan@amazon.co.uk>
MCP clients can now use \<Rightarrow> and ⇒ interchangeably in pattern, old_str, new_str, and content parameters. Whitespace differences (extra spaces, tabs) are also tolerated during matching. Signed-off-by: Hanno Becker <beckphan@amazon.co.uk>
Signed-off-by: Hanno Becker <beckphan@amazon.co.uk>
Without timing data from REPL steps, agents have a hard time optimizing proofs for performance. Signed-off-by: Hanno Becker <beckphan@amazon.co.uk>
LLM-based MCP clients to I/R are sensitive to the amount of information the receive, but the I/R output contains a lot of noise, including: - Every evaluation is suffixed with `val it = (): unit` - 'Duplicate simpset' warnings This commit adds an extensible noise filter to I/R to increase the information density of the response to the LLM. Signed-off-by: Hanno Becker <beckphan@amazon.co.uk>
Timeout was a single global Synchronized.var shared across all REPLs, so one MCP client changing it affected every other concurrent session. Move it to a per-REPL field: init/init_from_document default to 10s, fork inherits from parent, and Ir.timeout now takes a REPL ID. Signed-off-by: Hanno Becker <beckphan@amazon.co.uk>
Symptom: when multiple MCP clients maintain persistent TCP connections to I/R (e.g. repl.py's default pool of 5 + 1 console = 6 connections), Isabelle/jEdit becomes sluggish or unresponsive — PIDE proof checking and interactive editing stall even though the REPL connections are idle. Analysis: I/R's TCP server used Future.fork (tcp_handler.ML) and Future.forks (ml_repl.ML) to spawn connection handler and accept loop threads. Isabelle's Future system is a fixed-size worker thread pool (typically max_threads = 8), not raw OS threads. Each Future.fork body that blocks on socket I/O permanently occupies a worker slot — the worker cannot return to the pool to pick up other tasks. With 6 persistent connections + 1 accept loop, 7 of 8 worker threads are parked on blocking reads or Synchronized.guarded_access, leaving a single worker for all PIDE tasks and I/R eval futures combined. The eval futures themselves were already correctly prioritized at ir_pri = ~1 (below PIDE's >= 0), but this was moot when no workers were available to execute them. Fix: replace Future.fork/Future.forks with Isabelle_Thread.fork for all I/O-bound threads (TCP connection handlers and accept loop). Isabelle_Thread creates standalone OS threads outside the Future worker pool — the standard Isabelle idiom for I/O threads (cf. Message_Channel.make in message_channel.ML). Connection threads can block on socket I/O indefinitely without reducing the pool's capacity. The full worker pool remains available for PIDE and I/R eval futures. Output routing is unaffected: connection handlers use direct Message_Channel.message calls (not Private_Output), while eval futures still run in the worker pool with correct Future group ancestry (ir_group → conn_group → cmd_group) for find_connection_message(). Added prominent warnings at is_server_context(), connection_serve(), and eval_in_group() documenting that Private_Output wrappers (writeln, warning, etc.) only route correctly on Future worker threads — code on connection Isabelle_Threads must use the direct `message` function. Signed-off-by: Hanno Becker <beckphan@amazon.co.uk>
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
1 participant
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.
Symptom: Sessions with multiple
theoriesblocks in ROOT failed when built remotely via I/P, with repeated "Connection refused: failed to open socket 127.0.0.1:" errors. The build completed all theory loading but exited with rc=1.Issue: When proxying a build to a remote machine, the proxy rewrites bash_process_address in PIDE messages so that the remote ML process connects to a remote Bash.Server instead of the local one. The build_session message carries a separate Options object per theory group, each containing bash_process_address. The rewrite only replaced the first occurrence (count=1), leaving subsequent groups with the unrewritten local address.
Fix: Remove the count limit from replace() for both bash_process_address and bash_process_password, so all theory groups are rewritten.
Issue #, if available:
Description of changes:
By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.