Proto commits in liljencrantz/crush

These 40 commits are when the Protocol Buffers files have changed:

2026-09-12

Join non-last pipeline stage threads in Job::eval(), fixing swallowed failures Job::eval() only ever tracked and joined the last stage of a pipeline's thread. A non-last stage dispatched to a worker thread (can_block=true, the default for most builtins) had its result -- success or failure -- discarded outright: e.g. `... | uniq | echo` with uniq failing (not the pipeline's last stage) silently vanished instead of aborting the script. Fixed with two pieces: - CrushError::is_send_disconnected(), matching only SendError -- crossbeam's SendError only ever means "every receiver was dropped", unlike the existing is_disconnected() (which also treats RecvError as benign for an unrelated reason: next_row() uses it to mean ordinary end-of-stream). - Job::eval() now joins every non-last stage's thread after the last stage finishes, propagating a genuine failure as the job's own but ignoring one where is_send_disconnected() (e.g. head/take truncating a stream early). This alone deadlocked `&`-backgrounded pipelines, since `&` desugared to a literal trailing `bg` stage whose entire point is to not wait for upstream to drain. Rather than detecting "is the last stage bg" at runtime (rejected: name matching breaks under shadowing; Arc-identity matching needs the command position resolved twice), `background` is now a real field on Job/JobNode set directly by the grammar's existing `&`-recognizing production. Job::eval() acts on it directly: dispatch the real last stage, register its output receiver for `fg`, return immediately -- none of the new joining logic runs for a background job at all. `bg` (the builtin) is now unrelated to job creation entirely; repurposed to resume an already-paused job in the background, parallel to crush:resume. pup (de)serialization of a Job carries the new field too, so it survives round trips (e.g. remote:exec of a closure containing a backgrounded job). See test-gaps.md for the full story, including a documentation bug found in passing (bg's own example used an infinite seq by mistake) and a rare, still-unresolved full-suite-only flake in tests/pipe2.crush/pipe3.crush that two concrete hypotheses (ThreadId reuse, shared Scope::is_stopped) were ruled out for with direct evidence, not just reasoning. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

2026-08-28

Misc tweaks

grpc: Finish out incomplete parts

2026-08-26

Merge conflicts

2026-08-18

Add test grpc service

2026-07-16

Misc

2025-07-24

Better source code tracking Stack traces, command names and location tracking in error messages, etc.

The documentation is generated from this commit.

2025-07-11

one_of type, including support in closure signatures

2025-07-07

Better completion argument handling

2025-06-23

multi-command command substitutions

2025-06-12

Rethink documentation of closures

2024-08-08

=Misc

2024-08-06

Return command

2022-10-01

Fix missing column format serialization

2022-09-28

Correctly handle single vs double dash switches to cmd

2022-09-24

Drop non-working path join syntax

2022-09-15

New commands New commands for viewing history and for filtering text via globs and regexps

2022-09-12

Remove some dead code

Drop symbol value type, minor code cleanup

2022-09-05

AST/parsing cleanup

2020-10-05

Add short description to namespaces

2020-09-22

Add table_input_stream:pipe method Many to many connections between input and output of commands.

2020-09-13

Track location of arguments, start using it in error reporting

2020-09-07

Location tracking in parser Huge complicated change to start keeping track of mapping from source line to a parsed command. This is useful both for vastly superior error reporting and also to make tab completion possible.

2020-08-28

Make pup encoding slightly more efficient

Bump dependencies, update time parsing, add time tests

2020-07-28

Better API for duration creation

2020-07-05

Spelling

2020-04-13

Serialization should be feature complete

2020-04-12

Finish up closure serialization Serialization and deserialization of closures should now be feature complete.

2020-04-11

Most of closure serialization working

2020-04-09

Bound ethods are now serializable

2020-04-06

Add most of the proto work needed to serialize closures.

2020-04-05

Support serialization and deserialization of builtin commands

2020-04-04

Serialize struct type

2020-04-03

Serialize more types

Split up serializer code in multiple files

Flesh out serialization a bit

Move serialization to separate module

Start working on value deserialization