search

epage / clapng

A full featured, fast Command Line Argument Parser for Rust
docs.rs/clap
Apache License 2.0
0 stars 0 forks source link

[modular] Pull out a clap_lex #228

Open epage opened 2 years ago

epage commented 2 years ago

Issue by epage Tuesday Oct 19, 2021 at 15:42 GMT Originally opened as https://github.com/clap-rs/clap/issues/2915

Please complete the following tasks

Clap Version

master

Describe your use case

We are looking to modular clap

As previously discussed at https://github.com/clap-rs/clap/discussions/2615, one part of this would be pulling out a clap_lex crate, similar to lexopt.

Describe the solution you'd like

Clap is a wrapper around clap_lex

Alternatives, if applicable

Expand lexopt for our needs (probably easier to do an extraction refactor)

Additional Context

No response

epage commented 2 years ago

Comment by pksunkara Tuesday Oct 19, 2021 at 19:21 GMT

I agree with the author of lexopt in that discussion. I think there are quite a few scenarios clap is currently used for where a CFG parser would not work. (To give more context, lexer is used in CFG parsers and is useless in PEG parsers).

From all the examples of clap usage I have seen for the past 2 years, I am confident that we need to keep this as a PEG parser. We can make this PEG parser modular by implementing the plugins proposal in #2832.

I would nominate to close this issue (take your time to do research/experimentation) if needed.

epage commented 2 years ago

Comment by epage Tuesday Oct 19, 2021 at 19:32 GMT

I agree with the author of lexopt in that discussion. I think there are quite a few scenarios clap is currently used for where a CFG parser would not work. (To give more context, lexer is used in CFG parsers and is useless in PEG parsers).

The lexopt author didn't saw it was impossible, just said lexopt is insufficient. They went on to say:

I like the idea of clap_lexer, though, so I'll follow this discussion with interest.

kbnapp also thought it was possible

Very cool, I like the API a lot. I keep wishing I had more time on my hands to make something analogous to this as an internal clap API that all the clap features are built on top of which would allow using cargo features to aggressively trim down clap's features in an opt in/out manner (and thus be able to get that sweet small binary size).

Could you give a concrete example of why lexopts general approach is incompatible with clap?

We can make this PEG parser modular by implementing the plugins proposal in #2832.

A future experiment I want to do is something like clap_derive that directly populates the struct, without using the builder API.

In general, I would like to raise the bar for all of these non-clap parsers so we can have discussions beyond "does it support non-utf8?" :)

These are what I would like to see unlocked by decoupling our lowest level parsing logic from all of the clap's policy. https://github.com/clap-rs/clap/discussions/2832 is insufficient for this.

I'm also hopeful this will help us reduce bloat, directly and indirectly by

epage commented 2 years ago

Comment by pksunkara Wednesday Oct 20, 2021 at 14:58 GMT

I think a simple example here would be -1. Would you parse it as a short option or a value by the lexer? There are other ambiguous things like this that would indicate that this is not a context free grammar. And lexers are only useful in CFG.

epage commented 2 years ago

Comment by epage Thursday Oct 21, 2021 at 13:59 GMT

I get the feeling you are thinking we'd have a signature like

pub fn lex(args: &[OsString) -> Vec<Token>;

Is that correct?

If so, then maybe we aren't using terms correctly but that isn't what lexopt does which is the base of my suggestion (both in name and in functionality).

lexopt processes one token at a time, driven by the caller. A simple call to next() will auto-detect the token type but you can tell it the next token should be a value by calling value(). We could even have a peek_next() to handle some of our precedence cases between values and args or values and subcommands.

This leaves it to the caller to handle what -1 should be, along with whether -svalue is -s value or -s -v -a -l -u -e.

epage commented 2 years ago

Comment by pksunkara Wednesday Oct 27, 2021 at 02:29 GMT

I understand what you are saying now. Let's agree to not call this lex anywhere because it gives off the wrong impression.

Once this is implemented, do you envision this to be used by the plugins?

epage commented 2 years ago

Comment by epage Wednesday Oct 27, 2021 at 17:54 GMT

Let's agree to not call this lex anywhere because it gives off the wrong impression.

So far I've seen little confusion over lexopts name and suspect it helped people get the intent. Now, if you have an idea for a better name, I'm open.

Once this is implemented, do you envision this to be used by the plugins?

I've not seen a case where a plugin would have access to this. The benefits I see to this work are

epage commented 2 years ago

Comment by blyxxyz Wednesday Oct 27, 2021 at 18:30 GMT

Provide a solid base of up and coming argument parser to build on top of, rather than everyone reinventing the wheel, but poorly

It seems you could end up with something tightly coupled to clap if it needs to know about settings like AppSettings::AllowNegativeNumbers. And I don't know how clap implements features like pointing out that while some option is invalid here it would be valid if you put it there after the subcommand, but that also seems like it could get in the way of being generic.

Clap is large and ambitious. If some third-party generic basis already existed then I wouldn't expect clap to use it because it's working at a scale where a bespoke solution is worth the effort. It's like how rustc and GCC use handwritten parsers instead of parser generators.

A basis shared by multiple parsers could be a good idea, but I would guess that extracting it from clap isn't the way to go. (Then again, I know almost nothing about clap's internals.)

epage commented 2 years ago

Comment by epage Wednesday Oct 27, 2021 at 21:19 GMT

My hope is that it doesn't need to know about any settings like that, those would all be driven by the caller. The main risk is that this doesn't simplify our code but makes it more complicated. If it doesn't work then we've learned something.

epage commented 2 years ago

Comment by pksunkara Wednesday Oct 27, 2021 at 22:21 GMT

So far I've seen little confusion over lexopts name and suspect it helped people get the intent.

This library may also be useful if a lot of control is desired, like when the exact argument order matters or not all options are known ahead of time. It could be considered more of a lexer than a parser.

It was more of a general idea, but that library is not lexer. A better name would be clap_parser or clap_parse_stream etc..

I've not seen a case where a plugin would have access to this.

My hope is that it doesn't need to know about any settings like that, those would all be driven by the caller.

And I envision that the callers here will be plugins.

epage commented 2 years ago

Comment by epage Wednesday Oct 27, 2021 at 23:35 GMT

And I envision that the callers here will be plugins.

I'm not aware of this fitting in with any of our existing plugin talk (value validation currently being the most concrete).

epage commented 2 years ago

Comment by pksunkara Thursday Oct 28, 2021 at 00:20 GMT

We discussed that in https://github.com/clap-rs/clap/discussions/2832#discussioncomment-1465059.

epage commented 2 years ago

Comment by epage Friday Oct 29, 2021 at 16:37 GMT

We discussed that in #2832 (comment).

There was not enough detail on that for me to know what its referring to or how it ties into this

So far I've seen little confusion over lexopts name and suspect it helped people get the intent.

This library may also be useful if a lot of control is desired, like when the exact argument order matters or not all options are known ahead of time. It could be considered more of a lexer than a parser.

It was more of a general idea, but that library is not lexer. A better name would be clap_parser or clap_parse_stream etc..

All it does is tokenization. All of the policy of how to parse those tokens is up to the caller. That sounds closer to a lexer in name than anything else and calling it a parser would be overselling it. Also, the term "parser" is being used in so many places in clap in different ways, its good for us to look for more specific terms.

epage commented 2 years ago

Comment by pksunkara Friday Oct 29, 2021 at 16:52 GMT

All it does is tokenization.

But we are not planning to do tokenization. We are planning to let them read the items one by one without us categorizing them. That's it. And that's called a parse stream IIUC. Ex: syn uses the same word too.

Lexopt does actually categorize those args (and thus tokenize) but as @blyxxyz pointed out, it doesn't handle the corner cases that clap does.

epage commented 2 years ago

Comment by epage Friday Oct 29, 2021 at 17:22 GMT

At least from my searching, syn is the only user of that term. I did see "Token Stream" used in a similar manner.

We are planning to let them read the items one by one without us categorizing them

I see what we create being relatively similar to lexopt except it will have peek functionality.

epage commented 2 years ago

Comment by pksunkara Friday Oct 29, 2021 at 17:42 GMT

I see what we create being relatively similar to lexopt except it will have peek functionality.

I am not sure how that would work with those corner cases (ex: negative values, multiple values) since we definitely need to leave up the categorization to individual parsers. But I guess it's a design for later.

I realise that we are just bikeshedding on the name of this module at this point. But yeah, I felt that the term lex would be too confusing since it's not CFG.