Elixir v1.6.0-rc.0 released

Elixir News & Announcements Elixir News
22

@josevalim: Done.
For all interested more info in #7147: Could not call defguard check inside defguard. issue.

Update: Bug confirmed and fixed in 6aadfba586b22b22550e9d839a3c89d27224579b commit.

4 Likes
23

@josevalim: Sorry for ping you again, but I don’t understand why I got same warning (I corrected my guards).

Full code

defmodule Example do
  defguardp is_rgb_integer(tuple, index) when elem(tuple, index) in 0..255

  defguardp is_rgb_base(rgb)
            when is_rgb_integer(rgb, 0) and is_rgb_integer(rgb, 1) and is_rgb_integer(rgb, 2)

  defguard is_css_alpha(alpha) when alpha >= 0 and alpha <= 1

  defguardp is_rba_css_alhpa(tuple) when tuple |> elem(3) |> is_css_alpha()

  defguard is_rgba(rgba)
           when is_tuple(rgba) and tuple_size(rgba) == 4 and is_rgb_base(rgba) and
                  is_rba_css_alhpa(rgba)

  defguard is_rgb(rgba)
           when is_tuple(rgba) and
                  (tuple_size(rgba) == 3 or (tuple_size(rgba) == 4 and is_rba_css_alhpa(rgba))) and
                  is_rgb_base(rgba)
end

My changes were to make some guards private and don’t called twice, but I’m getting this: warning: this expression will fail with ArgumentError

Minimal example

I wrote smaller example that generates same error:

defmodule Example do
  defguard test_me(rgba)
           when is_tuple(rgba) and
                    ( # comment this
                      tuple_size(rgba) == 4
                      and elem(rgba, 3) == :ok
                    ) # comment this
end
# and in another file:
require Example
Example.test_me({1, 2})

Now try to comment lines with ( and ) characters (as in comments) and recompile project. Warning will not appear now.

Is it a bug too?

24

The warning itself is not a bug because the compiler is being smart and saying: “this will always fail”. That happens because you are passing a literal and the compiler can figure it all out. If you move the tuple to a variable instead of passing it directly to the guard, then the compiler should stay quiet.

I would say the warning in the last example is wrong though, it won’t “fail with ArgumentError”, but it will “always return false”. Although the wrong messaging is a bug in Erlang, not in Elixir.

25

ok, makes sense - really thanks for explanation

26

@josevalim Is there planned support for a max column width option in the formatter? Most of my codebases wrap at 80 chars, but the formatter will unwrap anything between 80 and 100 chars. I can restructure most things to keep the limit, but seem to be out of luck for long aliases and functions with guard clauses.

27

It is already there, please check the docs.

28

Sorry, guess I totally missed it. :tada: :purple_heart:

29

Which page? There’s nothing about it on https://hexdocs.pm/mix/master/Mix.Tasks.Format.html, is there?

30

Please take a look at: Code.format_string!/2
Parsed options are applied directly to it.

See:

  1. format_file/3 at format.ex:129
  2. Code.format_string!/2 at format.ex: 281

Docs says:

Formatting is done with the Code.format_string!/2 function. A .formatter.exs file can also be defined for customizing input files and the formatter itself.

So anyone reading docs should at least check it, but I don’t found that parsed arguments are passed directly to it!
To be honest there is a section for arguments where no length is described and next section called .formatter.exs describes extra arguments. Keeping in mind that there is no info that arguments are passed to linked function it could be wrong interpreted like:

There are no more options other than described, however .formatter.exs supports few more. Unfortunately there are also no such option supported there.`

Reading this first time I through similarly. I was need to read source of this task to be sure.

btw. @josevalim
I found that this link for docs in v1.7.0-dev points towards v1.5.3 version which is big mistake, because this version does not have implemented linked function!

31

The link that doesn’t work is a lesser problem than options for mix format not being listed on the one page you would expect them, but yes, the link working exacerbates that considerably.

32

Author of this probably wanted to say that these options are task-specific only. There is no need to describe same options in multiple places. The real problem is text (and it’s location) that links to rest of options - it could be missed.
I think that sentence with this link should be a little changed and moved to Options section, so it will be easier to find it.

33

If the docs are not clear, please send a PR.

The issue with us not linking to the proper Elixir version has been fixed on ex_doc master.

Thanks!

1 Like
34

There is. mix format is user-facing and it’s obviously where people will be looking for “How do I configure it?”. Any options you can use should be availabe in the CLI with mix help format as that’s likely what people will check first. I’d never bother actually checking the online docs unless someone told me something I couldn’t see with mix help format was available.

If we don’t want to duplicate text, we should at least make function documentation embeddable on other docs pages so that this kind of indirection in docs doesn’t hide useful information, in my opinion.

35

This is a good idea, but I think there is currently no implementation for it - at least I don’t know it.
I have created #7150 PR to make options part more clear, so you will see that you need to check docs of this function in order to get all formatting options and with it you will not have a need to check online docs.

1 Like
36

cough

Really though my library is purposefully incomplete and a hack (I redefine def and so forth). It is something that should be done properly in the elixir core, but they’ve stated this feature is not planned due to various edge cases that are hard to work around.

1 Like
37

@OvermindDL1: Nice, but.

  1. no suport for private guards
  2. when calling guard directly I got AST instead of boolean
  3. at compilation time I got lots of AST in console
  4. no support for remote calls (nested guard calls)
  5. no support for multiple guards with same name and args i.e. pattern matching (only first is used)

Do you have a time to change it in your library?

38

How long are release candidates usually take to release?

39

Depends on few things. This release introduce lots of really usable changes that I would like to test, so I believe that it requires more time for testing them. Last time I tested new guards and rest I don’t even tried yet - I don’t have a time for it, so some other people probably too. Next things I will test tomorrow.

I think that it have a chance to be ready between 2nd and 3rd week of January. At least I plan to finish testing in this or next week and remember that it’s only my plans. I believe that some people that want to help could not even start work with it - job, family and rest time. For some people real work could start at 1st or 2nd January.

1 Like
40

Thanks Eiji

41

It depends on how much feedback we get and how much of this feedback is positive. We will probably do another RC around 4th or 5th January, and the final version at most a week after that (assuming that there are no reports).