pitfall - IFS= read x doesn't set x (read --raw-line or for <> loop instead) about oil

[SylvAbdomen]

Interesting! Thanks for looking in to this. I didn't know about shvar. For now I can work around the issue with a loop like this.

proc p {
	var line = ""
	while true {
		try { shvar IFS= { read -r line } }
		if (_status !== 0) {
			break
		}
		write $line
	}
}

I just recently started learning YSH, and it's been great so far. Love the project!

[andychu]

Hmmm I just reproduced this ...

Let me look into this, thanks for the clear report!

[andychu]

Hm funny that it works in proc without IFS= (though removing IFS= does change the behavior, even for one read arg!)

---

But I think there's still a bug here. I think it is probably related to lack of dynamic scope in procs

Our replacement for dynamic scope is shvar IFS= { read -r line }, but that is actually problematic in loops, since it's a command

I'll look into it a little more ... hm

[andychu]

OK wow, now I see the problem...

It's because

1. IFS= anycommand creates a new stack frame in shell. That's how "temp bindings" are implemented in all shells

2. In OSH shell functions, read -r line uses dynamic scope. This means it often creates a global variable, unless you declare with local or var inside the function

$ f() { read -r zz; echo "inside f: $zz"; }; f <<< 'stdin'; echo "global: $zz"
inside f: stdin
global: stdin

Actually I was confused about this for a long time ... I thought that read -r line would logically modify the calling stack frame! But it doesn't -- it looks up the whole stack, and if nothing is there, it creates a global

3. YSH proc don't have dynamic scope. We got rid of that because it's unfamiliar to say Python and JS programmers

So basically YSH creates the line variable in the temp binding, and it gets immediately thrown away when read returns !!

---

So bottom line:

I will add

• read --bytes-of-line as a synonym for IFS= read -r
  • rationale on Zulip (please join if interested) - https://oilshell.zulipchat.com/#narrow/stream/417617-help-wanted/topic/for.20loop.20that.20buffers.20lines
  • related deprecation related to buffering - https://www.oilshell.org/blog/2024/03/release-0.21.0.html#breaking-changes

And then let me think about what we can do about this IFS= read pitfall

It will also affect mapfile and any other builtins which set variables

---

Hm hm

[andychu]

I implemented

• read --raw-line for unbuffered read
• for line in <> { echo $line } for buffered read

So you don't have to use the old shell idioms

Documented here

9911231

I guess I will turn this bug into "do something about IFS= read -r` in YSH ... that is confusing

Maybe we need a lint rule or something

[andychu]

Related to

• #1938