and make note of performance issue. I think optimizing Hay Attr node
detection will make this m...

without span ID.

Remove usage of the 'span_id -1' pattern.

Thinking about how to further optimize the parser.

- Improve abbreviation
- Note that BracedVarSub is 56 bytes

It's a singleton class, not SimpleVarSub %Token, because

- word_part_t has word_part.Literal %T...

Might as well use the same mechanism for OSH here docs and YSH
multi-line strings.

The lossless...

We introduce Id.Lit_CharsWithoutPrefix, and add a special case in --tool
lossless-cat.

TODO: re...

So now we have some EAGER (Token, str) pairs like SimpleVarSub.

And some that use lexer.LazyStr...

Use lexer.LazyStr() for the variable, in arith evaluation.

For y_lhs

Fix build by removing TokenWithStr

[frontend refactor] Simplify re_repeat schema

Remove TokenWithStr. We can just use Token and t...

- Introduce lexer.LazyStr(tok)
- Remove direct usages of Token.tval

TODO:

- Some NameTok / Tok...

Just check tok.col == 0.

Remove Id.Ignored_LeadingSpace. We have a more elegant mechanism now....

This will make whitespace stripping and the lossles variant more
uniform.

We just have to test if tok.col == 0. Duh.

Improve function names.

Algorithm for stripping leading space doesn't depend on Token.tval

Change signature of whitespace stripping algorithm.

Instead of List[Token], just store the string.

This means we don't use Token.tval at runtime.

...

This method is useful for avoiding intermediate slices.

Remove lexer.DummyToken()

Optimize schema of re_repeat_t.

They are constants, so it belongs at parse time.

I think the old style was left over from when ...

Clean up re_t schema.

Add blame token to re.Primitive

Observation: we have 300+ extra processes that don't call execve()!
They're almost certainly her...

And

type -a ''

Also able to write out *-argv0.json files, for multiple processes.

It need...

Dumping histogram of argv0 as a JSON file

Still need to do it across multiple processes.

Issue #1864

Compare running time against bash and dash.

I see a big discrepancy in the perf profile vs. our...

It seems like this will work! We can create StackArray[T], which is
similar to List[T], but not...

This task runs oils-cpp-smoke, which has been unstable on Github
Actions.

https://github.com/actions/runner-images/issues/9491#issuecomment-1989718917

This is issue #1862

This is issue #1862

And fix typo in release-quality.md

Also need to switch machines for release - lenny is machine1, with
bloaty.

No more warnings now!

The code is clearer this way.

Also fix soil/worker.sh after renaming function.

[release] Add packaging guidelines.

Change precision of osh-runtime ratios. We should take about 1.05x as
long as bash on CPython c...

Well the optimization did what I intended, but it didn't move the
overall wall time as much

The array in test_strings.py was trivial, so it didn't catch this!

Added mylib.JoinBytes()

To speed up IFS and glob manipulation

Put oils-for-unix tarball first on the downloads list. We've translated
it!

Figured out why CPython configure runs in 1.1x the time as bash -- small
1-byte string objects! ...

Not conclusive. We might rely on 'perf' more?

[mycpp/examples] Document idea for simplifying cast pattern

We are making the "UP_val downcasting pattern" real in mycpp.

This fixes a failing test case in...

- _MakeAssignPair() no longer tickles it
- Disabling it no longer makes tests fail

Add failing ...

[doc/ref] Fix TOC title and intro

This moved the needle on a few benchmarks! Including the fib workload,
e.g. benchmarks2/gc-cac...

It only happens in str_switch().

In contrast, tagswitch() are often exhaustive.

It first dispatches on the string length, and then the value with the
runtime function str_equal...

Collect a flat list of cases.

- arg-parse -> parser
- default= and help= are named arguments
- Enforce required arguments in...

- Consistency
- Naming should be "start" and "end"
- startsWith() and endsWith() were th...

Should turn _allow_command_sub => _no_command_sub I think

Other minor updates

Job control with pipelines is a inherently racy.

Zulip thread:

https://oilshell.zulipchat....

Document this quirk.

This addresses issue #1822.

This is after the last change, which removed getpgid() calls.

We shouldn't access the Pipeline....

One of them causes the ESRCH bug, triggered by starship and others.

Repro:

$ bin/osh
osh...

In bash, the first part gets a DEBUG trap:

echo hi | cat

But here it doesn't:

{ echo ...

break / continue / etc.

Add spec test case.

And inline small functions.

We can reproduce issue #1822 reproduce the issue with our own code.

The problem is:

- A DEBUG ...

Make note of interleaved set -x output, even when shopt --set
xtrace_rich.

To import from vendor/

[refactor] Move jsontemplate.py to vendor/

It takes much longer to print than to parse. See snippet in shell
script.

Also added a test case for the ESRCH bug, but wasn't able to repro.

It seems like useful/missin...

No dice

[test] Delete old files, now that we've translated to C++

Also don't use pid_t as synonym for int.

I think there may be a race in the code with getpgid().

We're like bash/zsh, not mksh!

Update comments.