x/tools/gopls: no completion for function literal in parameter

[stamblerre]

Triggering completion here should offer a suggestion for a function literal.
The function call pictured below is https://pkg.go.dev/golang.org/x/build/maintner?tab=doc#Gerrit.ForeachProjectUnsorted, which takes a function as a parameter.

/cc @muirdm

[muirdm]

That func literal requires placeholders to be enabled since the function signature doesn't give a name for the parameter. gopls gives a placeholder for func(<g> *GerritProject) since it is just guessing on variable name.

If we change the placeholder option to only apply to function invocations, then this problem would go away.

[zikaeroh]

I've hit this a lot too, and didn't realize I'd need to have placeholders to do it (I don't like them, so have them disabled) so had been assuming it was a weird bug that only affected some code.

Would it be reasonable to insert with the parameter names as _ in the no-placeholder case? (Enabling placeholders outside of calls would work too, of course.)

[muirdm]

Would you dislike placeholders in this case as well, or do you mainly not want the function call placeholders?

The reason not to include "_" was because it is inconvenient/annoying to go back and update them. In general I feel like having to go "backwards" to change something is quite annoying, even vs having to type out the entire signature.

[zikaeroh]

I gave it a shot, re-enabling placeholders and checking this. Given it's a declaration, I don't mind the placeholders so much.

A small part (but definitely not all) of my issue with placeholders for normal calls is that it inserts "incorrect" code by inserting the type names (which then need to be replaced with values of that type). For decls, that's not the case, as it inserts say c context.Context as a parameter, which is valid code when declaring a function parameter.

This doesn't solve my other personal usability issues with placeholders, but I think it's fine in this case and better than inserting _ (since you get generated parameter names).

[gopherbot]

Change https://golang.org/cl/228104 mentions this issue: internal/lsp/source: limit effect of usePlaceholders option

[muirdm]

(but definitely not all) of my issue with placeholders

What else don't you like about placeholders?

[zikaeroh]

I'll try to explain it, but this is of course personal preference and years of workflow gained from previous Go tooling.

1. I'm just not used to it (this is the first experience I have with it, in any editor or language outside of pre-defined "snippets", which I also dislike and disable). A big blob of text appears that doesn't look like anything I'm going to write and my brain goes "okay time to delete this". Without placeholders, it just adds the () to the function -- which itself took a good while to get used to (and isn't 100% right in all cases, especially when I'm trying to work with functions as values and have to undo the parens).
2. In VS Code, the moment you move the cursor, the placeholder is "committed" and stays there, broken code and all. If I'm writing something and leave to go refer to what I'm calling, say go-to-def via moving the cursor and F12, accidentally click, etc, then I've got to do cleanup afterward. It's unlikely that I'm going to know exactly what I want to write in order to get it all right the fist time when tabbing through. And at the end of the day, I'm still going to have to type out the parameters. Placeholders are only saving me the commas, and costing me time when I make a mistake and break the workflow.
3. I find much more value in the signature help while I'm typing than I do any placeholder; it's the same info except it doesn't get in my way (with the above items) and is re-triggerable if it happens to disappear.
4. The "lame" reason that from the correctness POV, placeholders still put "invalid" code down (copy the signature, including types), so if I leave it then I'll end up with more red squiggles. Note that this is the case with almost every other placeholder, particularly struct literals where the placeholder is the type name (as the name has already been written to trigger the compltion).

So in comparison to your (100% reasonable) statement of "having to go backwards to change something is quite annoying", I find it annoying to have to undo what completion's done that I didn't like or "ask for", and the only benefit is not having to type a few commas.

Now, I know a lot of people like placeholders, and wouldn't want my workflow to break someone else's. I appreciate having the toggle so I can work happily.

I see that the above CL disables placeholders in other contexts as well. I'm not sure if I'm into that, for similar reasons as 1/2 above. I get things like:

Which end up distracting me in a similar fashion. The code isn't real, I have to clean it up, I'm just going to re-type it anyway, etc.

For function literals I think "forcing placeholders" (or a least filling things out automatically) makes sense as there is no way to end up with a usable function when the signature doesn't have names. It's certainly much better than not giving literal completion at all and leaving people confused as to why the (very cool) feature didn't work this time around.

But I'd be sad if the placeholders option stopped working for the other placeholders. It's much more likely that I'd just live with some function literals being mysteriously "broken" so I can save myself the trouble in the other contexts where I know it's going to slow me down. (EDIT: though now that I think about it, there just wouldn't be any way at all to disable those placeholders now, so I'd just be frustrated and leave my gopls old, or patch it myself... 🙁)

Can this change only be made for the specific "function literal without parameter names" case? Or, with placeholders disabled, just leave the auto-generated names?

[myitcv]

@zikaeroh very much agree.

We've covered this topic a few times, most recently in #37656 (comment). This issue seems to add weight to the idea that there needs to be different configuration options for the various types of placeholders that people might want:

1. no placeholders
2. a single placeholder that represents the cursor location
3. ...
4. what this issue is asking for
5. the full monty; all placeholders in all places

(In #37656 (comment) and elsewhere I've argued that option 2 could be implemented using multiple text edits as opposed to snippets, but that's an orthogonal concern).

[zikaeroh]

One extra thing I thought up after my comment (and numerous edits, sorry) is that I also view the "label" of the completion to signify the intent. If the completion says func(...) { }, as function literal completions do, it's clear to me that I'm about to generate something interesting. Same with the new if err != nil { ... } option (which I didn't realize had been locked behind placeholders until I read the aforementioned CL).

However, when I'm in a struct literal and the completion is Foo, I don't expect to end up with Foo: string,. The label said one thing, but I ended up with more than I agreed to.

I'd even go so far as to say that completions for function calls where the label is SomeFunc, I'd expect to end up with SomeFunc, but end up with SomeFunc() or with placeholders SomeFunc(huh string, foo int, vars ...interface{}). I'd advocate for the one with placeholders to have a label like SomeFunc() or SomeFunc(...) to indicate that it's about to create a call, and potentially even say that there should be completion items for both SomeFunc and SomeFunc(). But, those sorts of changes are out of scope for this function literal issue. 🙂

[muirdm]

Thanks! It's really helpful to know other people's thoughts.

I sympathize with the placeholder text being invalid Go - that bugs me too.

For function call placeholders, my editor does not present signature help very well, so I rely more on the placeholders. Other than that, I don't like them either.

The struct field placeholder in particular may not be that useful. Perhaps the solution there is to just remove the placeholder altogether. I agree it is strange to just jam in some informational message as source code.

That being said, I don't understand the "have to clean it up" or "have to undo" sentiment (for struct field placeholders at least). Normally I just start typing what I want to complete and it overwrites the placeholder, so there are no wasted keystrokes. Your scenario in 2) of course causes extra work, and that has bugged me for function call placeholders.

I'd advocate for the one with placeholders to have a label like SomeFunc()

I agree. The label should be clearer.

there should be completion items for both SomeFunc and SomeFunc()

Are you still running into issues using functions as values? Maybe it is due to an expected type of interface{}?

I don't have any good ideas for how to proceed with this change. Long term I would love it if we could eliminate all the controversial use of placeholders and just get rid of the usePlaceholders option, but that might not be possible.

[zikaeroh]

That being said, I don't understand the "have to clean it up" or "have to undo" sentiment (for struct field placeholders at least). Normally I just start typing what I want to complete and it overwrites the placeholder, so there are no wasted keystrokes.

A lot of the time, I probably don't know what I'm about to type as parameters, or I get distracted. As stated in my issue, if I "leave", that stuff gets leftover and I have to fix it. The only keystrokes I'm saving are the commas, and I have no trouble typing those myself. I get that your editor doesn't display signature help too well, but mine does and I don't like the rest of the implications of the placeholders.

Also note that at least in VS Code (I think you use emacs?), typing a comma does not advance the placeholders, only tab does. This is frustrating as in every other language (and in Go previously), I have to type the commas. For example:

I'd rather just type it out.

Are you still running into issues using functions as values? Maybe it is due to an expected type of interface{}?

Simple case, x := someFunction. There's no way to know if I intend to call this function or not. Most of the time, I do, but sometimes I don't. I'm not really expecting gopls to know what I want. The extra completion items with distinct labels may help here, but deleting () is way easier than deleting an entire signature's worth of text.

I've also seen gopls add () and such when I use functions as literals:

It's extremely irritating. With placeholders off, it's less frustrating (just deleting ()).

Long term I would love it if we could eliminate all the controversial use of placeholders and just get rid of the usePlaceholders option, but that might not be possible.

I don't see this ever being possible. I'd much rather things continue to be configurable. For this specific issue, I think leaving the auto-generated names is the most "correct" thing to do for "usePlaceholders": false given it leaves valid code without placeholders.

I think it's possible to come up with a range of options that would make this more customizable as @myitcv said above, though it may take a little defining. I haven't personally found a case (besides this bug) where I'd actually want the placeholders to warrant extra levels, however. @myitcv has a technical motivation from the standpoint of not wanting to handle snippets at all (I think), which is more than I need given I'm a VS Code user.

[zikaeroh]

I entirely skipped over your "for struct field placeholders at least" when writing my first paragraph, my mistake. I thought I hadn't expressed my original thought well enough but you were just narrowing it to the struct stuff. Oops...

Generally I view all of the placeholders as "the same", at least from the "annoys me" POV where I have to undo. This is probably due to my first bullet thing before where I'm not used to it, and I feel weirded out by the extra text I didn't request (my follow-up label point).