unexpected behavior from regexp substitution

[JeffMelton]

I haven't seen this behavior documented anywhere, so I suspect it's a bug, but I'll start here in case it's PEBKAC. I wanted to remove all leading zeroes from a string of integers, so out "0123" -> regexp 's/^0*//' seemed sensible. Documentation for regexp indicates that the ^ anchor for beginning of line is supported, but ^0 seems to be some kind of syntactic sugar (undocumented, perhaps?) for "very first history entry".

I was able to get what I needed with 's/\A0*//', but I figured this was worth asking about.

[lmorg]

You're right on both counts.

^ is used as a token for history variables. And you're right that it's undocumented.

It's also amongst some of Murex's oldest code: https://github.com/lmorg/murex/blob/master/shell/history/vars.go

Given how old the code is and how it's undocumented, I'm definitely open to revisiting / reworking this implementation.

One thought I'd been having lately was having this a feature of readline. Making it a command mode like vi's : command mode. So it could be invoked via [esc]:

[JeffMelton]

I don't have strong feelings about it one way or the other, I suppose. I don't have much use for that history feature myself, but I do lean heavily on regex anchors, and I hadn't been aware of the \A one until now. ^ and $ are carved into my muscle memory right next to Vi motion keybindings. 😄

I would say that until you've decided what you want to do with that feature the docs should probably be updated in the regexp section to point out the quirks around using ^, since that applies not only to the beginning-of-line anchor but also to character class negation. Is there a fancy quoting or escape approach that would avoid triggering that history feature, or is there a config flag for disabling it? I tried a single backslash, but once I found \A I quit playing with it.

[lmorg]

Presently there's no way to escape nor disable it. The only saving grace is that it only applies to stuff typed in the REPL. read doesn't use it and nor do shell scripts.

To be honest, I never completely liked the ^ pattern. It made sense back when Murex was in beta and I used a lot more "off-the-shelf" libraries. But now I've rewritten everything else, it's time for me to rethink this too.

Some of the functions, like ^s/find/replace/ definitely don't belong as prompt variables and should be integrated into readline instead.

[JeffMelton]

Ah, ok, thanks. 🙏🏽

[lmorg]

The history variables haven't been replaced, however a first draft of porting that functionality to readline is in the following branch: #907

Here, you can access the command palette via {ESC}: (like you would in vim too)

Supported commands are:

"Valid commands:":   "",
"  !!":              "Previous command line",
"  n":               "history item n (where 'n' is an integer)",
"  m/find/":         "match string (regexp)",
"  s/find/replace/": "substitute in current line (regexp)",

however more can and will follow