ergochat / readline Goto Github PK

CharLineStart, MetaBackward, etc. can all be lowercased, unexporting them.

term*.go are vendored forks, due to chzyer/readline#62 . The underlying issue is still present in x/term 0.6.0. It has something to do with this specifically:

https://github.com/golang/term/blob/0edf009663f942c41572b65d537ee2636e92180c/term_unix.go#L34

Context: golang/go#15625

I bisected this to 62ab2cf. There is now a race where the password prompt can be printed an extra time. It's reproducible with example/readline-pass-strength/readline-pass-strength.go:

$ ./readline-pass-strength
✔ ENT New Password: 
✗ ENT New Password: Your password was: sdjflaksdf

I'm considering a more extensive refactor of how config updates are done, but I want to get a firm diagnosis for this first to make sure I'm solving the root cause.

Tested on WezTerm, Windows Terminal (both use OpenConsole) and normal Command Prompt. WSL has normal expected Linux behavior.

Should be straightforward because we already accept Ctrl-A and Ctrl-E?

Hey, just happen to noticed this:

$ go run example/readline-demo/readline-demo.go
»
# now press ctrl-r to get the "bck-i-search" promtp, then press enter
» k-i-search:
# ends up with messed up prompt

Think i also managed to get the history search into some other weird states but can't really reproduce it now

The expandable ring buffer from #46 can replace container/list as the data structure for history. Just a nice-to-have.

Hey! hope your doing good! I noticed this patch for the z/OS fq port, merge?

https://github.com/ZOSOpenTools/fqport/blob/main/patches/readline-patch

As discussed here: wader/fq#940 this is a normal readline feature. I checked our v0.0.1 (last upstream release) and it does not work there, so it's not a regression, it was never implemented.

See #61, it's not clear that this is really useful for anything (although it also doesn't clutter the API very much).

Add documentation explaining what has changed from upstream and how to migrate.

We've accumulated a lot of cases that are likely to break:

1. Async output (use the async command in readline-demo)
2. #38
3. (Windows special cases?)

e.g. (*Instance).ReadlineCtx(ctx context.Context)

chzyer/readline#163 is still present in our fork

This code is extremely bogus:

Instead, resumption should be detected by handling SIGCONT. (Not sure about the story on Windows.)

Suggested by @tpodowd

What do the "clean" methods do that isn't accomplished by redrawing an empty prompt line?

Let's do this last, after other invasive refactors are complete, since it will introduce lots of potential merge conflicts.

Things like:

• Less refresh flicker on resize and editing
• Modern/Better windows ANSI support
• TODO: Undo support
• Lots of bug fixes
• Less and updated dependencies

We are trying to migrate from chzyer/readline to this fork since the former seems to be unmaintained.
But I'm facing a few issues.

github.com/chzyer/readline had a ClearScreen method that was quite convenient for us as we want to expose
users a way to clear the screen by typing clear. This package can still do that when pressing Ctrl+L but doesn't espose that as one easy method.

Could we get a ClearScreen method exposed, please?

Investigate https://github.com/charmbracelet/vhs for this: https://www.youtube.com/watch?v=dtVezRX8OEM

1. Hide (*Config).Init: this should not be exposed
2. NewFromConfig and SetConfig should copy the config. This prevents data races from clients attempting to modify the config from the outside.
3. Multiple copies of the config are stored in different structures. All config access should be consolidated into one canonical location that can be updated atomically
4. It will still be necessary to propagate some changes into subsystems when the config is updated; this should be made data-race-free
5. Remove Config.opHistory and Config.opSearch (these should be members of Operation only)
6. (*Instance).ReadPasswordWithConfig has weird implications, for example, it can change the stdio handles back to the defaults. This API needs reevaluation.

This race is still present:

chzyer/readline#219

Ctrl-R writes to the search mode status, wrapWriter reads it on asynchronous writes, and these are not necessarily synchronized.

It's unclear that FillableStdin is actually usable for any purpose that ReadlineWithDefault can't handle:

This is looking better:

https://pkg.go.dev/github.com/ergochat/[email protected]

but there's still a lot of cruft. For example:

1. Listener should probably become a function pointer type rather than an interface, eliminating the need for DumpListener. See #6 for discussion of the listener API
2. The relationship between PrefixCompleter, PrefixCompleterInterface, and functions like Do and Print is confused
3. Instance has too many methods
4. Painter should be a function pointer type, not an interface type
5. TabCompleter should probably be hidden
6. What are SegmentComplete and SegmentCompleter?