Open lpil opened 8 months ago
I'm a fan of whatis
.
I like whatis
but the fact that it's not what_is
would be painful to me!
I vote peek
.
One annoyance is that the anything-to-string code is in the standard library, not in core. Perhaps it could gracefully degrade if stdlib isn't available.
What behavior would it degrade to?
Big fan of whatis
as well
One annoyance is that the anything-to-string code is in the standard library, not in core. Perhaps it could gracefully degrade if stdlib isn't available.
What behavior would it degrade to?
Printing in the syntax of the native target. JS or Erlang
I'd like both what_is
or peek
, even whatis
(but I prefer the snake case version)
Where could it be used? Would it work like debug returning its rhs?
peek peek foo
// print foo twice?
bar(peek foo)
// print foo?
foo
|> peek bar
// print bar?
Having it return the value sounds good. We'd need to decide if it has any special pipes support as I suspect folks will want to pipe into it.
Just call it glimpse
and be done with it :wink:
👀 value
No seriously though I would prefer it just be called debug or dbg or debug_print or something to that effect. I wouldn’t mind a naming conflict and deprecating io.debug.
I think the ideal syntax would be for it to work in a pipe and for keyword value
to be how it’s called
Edit: I no longer think this syntax is a good idea, I would much prefer something like a global function.
We're not breaking the stdlib, and there's lots of other contexts in which the word debug is more appropriate and already taken, such as logging libraries.
I don't think debug
is as good a name for printing as it would be for use in a debugger too.
pry
is the only one not previously said that comes to mind for me
spy
because it makes me feel cool
peek
and dump
both are single syllable and have no underscores and are short enough.? exp
:? foo
|> ? bar
|> ? quux(? batz)
I like it :)
pry
is the only one not previously said that comes to mind for me
but are we really prying... into the vm runtime value?
My personal favorites are dump
and ? expr
. whatis
and what_is
are both slower to type and to say (even mentally), so I'm not as big a fan of them.
I'm just a silent reader on discord, but I'll chime in for bike shedding :laughing:
What about trace
?
Apparently that even has some precedent in ActionScript: https://open-flash.github.io/mirrors/as2-language-reference/global_functions.html#trace()
What about
trace
?
Seems like we don't want to do anything that would overlap w/proper logging (trace
, debug
, info
, warn
, error
).
?
has a lot of potential power as a single character operator, I would hate to see it wasted on this.
What about echo
as the name?
Does it have to be a keyword? Something that worked more like a global function would be a lot more composable w/gleam syntax in my opinion, and you wouldn't have to adopt a whole new style of control flow for a single keyword. Tbh this feels so clearly like a function I think it would be extremely strange to have it be a keyword as described above.
echo
is actually a really good suggestion! I like that a lot.
Does it have to be a keyword? Something that worked more like a global function would be a lot more composable w/gleam syntax in my opinion, and you wouldn't have to adopt a whole new style of control flow for a single keyword. Tbh this feels so clearly like a function I think it would be extremely strange to have it be a keyword as described above.
It is to do things that a function cannot do. I'm not sure making it look a bit like a function would be beneficial given it couldn't be one.
echo
is actually a really good suggestion! I like that a lot.Does it have to be a keyword? Something that worked more like a global function would be a lot more composable w/gleam syntax in my opinion, and you wouldn't have to adopt a whole new style of control flow for a single keyword. Tbh this feels so clearly like a function I think it would be extremely strange to have it be a keyword as described above.
It is to do things that a function cannot do. I'm not sure making it look a bit like a function would be beneficial given it couldn't be one.
that's the reason I like ?
as it cannot be confused...
What would the semantic explanation be for echo versus peek or pry or dump?
What would the semantic explanation be for echo versus peek or pry or dump?
No semantic difference here, just about how they feel, look, and any connotations from other languages.
hm, echo would write a string to the output buffer in php. which other languages use echo?
echo foo
|> echo bar
|> echo quux(echo batz)
I think the question mark with a space is a good thing, ... a word kind of pollutes the screen/attention and makes it harder to follow the code with the debug printing included.
Ah yeah "looks like a function but isn't a function" is actually a bad outcome I suppose 🤔
If I hadn’t been poisoned by previous language experience I would think value?
would be a quick easy common sense way to have a built in debugging feature, but having seen all the powerful ? syntax in other languages it would feel like such a demotion for the ?.
I wonder if we'd regret using ? here down the line when we have an idea for a cool new feature that ? Would be perfect for.
I doubt we'll ever use ?
for anything. It's very easy to miss, not very Gleamy.
I would like thing?
but ? thing
seems too weird to me. I don't see how it fits in Gleam! I really feel like it breaks the flow somehow!
But I should add that a keyword seems more Gleamy. I think pry
or peek
are great.
Could gleam publish fail if there are debug commands in the code?
Yes! That's one of the things covered above
Goals:
- Prints location so the programmer knows where the logging is coming from
- Prints optional extra context, like a tag
- Does not require an import
- Warns so you don't forget it by mistake
- Blocks
gleam publish
so you can't publish code that uses it
Let's keep in mind goal 2. Including a "tag" in the output in order to label the thing being printed can be extremely helpful in aiding debugging.
I don't think any of the syntax proposals so far include such tagging.
How about something like the following?
value
|> process
|> :tee "my tag"
|> more_processing
This potentially does the following:
:
prefix to distinguish as not a function (since it's not a valid function name) -- see alternative suggestions belowtee
command, which passes its input through unchanged to its output, while also spitting it out to a file (in this case, spitting it out to either stdout or stderr, which is perhaps another question to consider)"my tag: x"
(where x
is the value returned from process
) to either stdout or stderrI think the question of where to write the output (stdout or stderr) is perhaps another consideration. I would lean towards stderr so that injecting :tee
doesn't potentially interfere with the program itself, which might be intended to be used in command pipeline (like the tee
command itself). If it were to write to stdout, it might interfere with proper operation of the program that we might be trying to debug.
Alternative syntaxes for the :tee
shown above:
|> :tee "my tag"
|# "my tag"
|> #"my tag"
|> ("my tag")
|> #my tag#
We could also reuse the "as" used by todo and panic:
foo
|> bar
|> echo as "tag"
|> baz
It would be a bit more awkward when used as a function though:
// maybe? I have no idea how it should look
echo foo as "tag"
I think it should print to stderr. If required later gleam.toml
could make this behaviour configurable.
For words, short 1-word-1-syllable dump
, pry
, peek
, show
are those I'd prefer.
If we need tagging for the question mark, if we do prefixing, it could look like this:
// Regular
? var as "tag1"
// Piped
? foo as "tag2"
|> ? bar as "tag3"
|> ? quux(? batz as "tag4") as "tag5"
...or if it is suffixed like this:
// Regular
var ? "tag1"
// Piped
foo ? "tag2"
|> bar ? "tag3"
|> quux(batz ? "tag4") ? "tag5"
inspect foo
Inspect is not an option, it is already taken. We have gone with echo, the discussion is the design now.
echo thing as "thing"
seems reasonable to me.
@lpil, as mentioned above, I'd suggest output go to stderr, not stdout, to avoid potential interference with intentional program output. Does that make sense?
Definitely. This is also what io.debug does today.
If this doesn't make it into 1.0 would it be sensible to reserve echo
?
If this doesn't make it into 1.0 would it be sensible to reserve
echo
?
It was reserved in v0.34.0
Goals:
gleam publish
so you can't publish code that uses itWhat should the syntax be? Update: The keyword is echo.
One annoyance is that the anything-to-string code is in the standard library, not in core. Perhaps it could gracefully degrade if stdlib isn't available.