Layout rules for if then else - Haskell-Cafe - Haskell.org

newer
Bad recursion? When can something...

Layout rules for if then else

older
Monadic/sequential migration...

Alan & Kim Zimmerman

10 Mar 2015 10 Mar '15

12:01 p.m.

I am working on ghc-exactprint, and need to flag points where layout must be preserved, and considering the `if` statement. My understanding of the layout rules for if then else is that the `then` has to start more to the right than the `if`. Using GHC 7.10 RC2, ghci cheerfully loads the following ``` f = if True then 1 else 2 ``` Is this valid? Have the rules been relaxed? Alan

Attachments:

attachment.html (text/html — 506 bytes)

Reply

Sign in to reply online Use email software

Show replies by date

Vo Minh Thu

10 Mar 10 Mar

12:04 p.m.

What you state is true within a `do` expression. What you wrote is a single if_then_else_ expression on the rhs of the function definition. Cheers, Thu 2015-03-10 17:01 GMT+01:00 Alan & Kim Zimmerman :

I am working on ghc-exactprint, and need to flag points where layout must be preserved, and considering the `if` statement.

My understanding of the layout rules for if then else is that the `then` has to start more to the right than the `if`.

Using GHC 7.10 RC2, ghci cheerfully loads the following

``` f = if True then 1 else 2

```

Is this valid? Have the rules been relaxed?

Alan

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe

Reply

Sign in to reply online Use email software

Brandon Allbery

12:06 p.m.

On Tue, Mar 10, 2015 at 12:01 PM, Alan & Kim Zimmerman wrote:

I am working on ghc-exactprint, and need to flag points where layout must be preserved, and considering the `if` statement.

My understanding of the layout rules for if then else is that the `then` has to start more to the right than the `if`.

I believe you are looking for DoAndIfThenElse. -- brandon s allbery kf8nh sine nomine associates allbery.b@gmail.com ballbery@sinenomine.net unix, openafs, kerberos, infrastructure, xmonad http://sinenomine.net

Reply

Sign in to reply online Use email software

Alan & Kim Zimmerman

12:25 p.m.

Ok, adding a do does make a difference. More complexity. Thanks Alan On Tue, Mar 10, 2015 at 6:06 PM, Brandon Allbery wrote:

On Tue, Mar 10, 2015 at 12:01 PM, Alan & Kim Zimmerman < alan.zimm@gmail.com> wrote:

...
I am working on ghc-exactprint, and need to flag points where layout must be preserved, and considering the `if` statement.

My understanding of the layout rules for if then else is that the `then` has to start more to the right than the `if`.

I believe you are looking for DoAndIfThenElse.

-- brandon s allbery kf8nh sine nomine associates allbery.b@gmail.com ballbery@sinenomine.net unix, openafs, kerberos, infrastructure, xmonad http://sinenomine.net

Reply

Sign in to reply online Use email software

Andrew Gibiansky

2:05 p.m.

Just to be clear, I *think* layout rules don't apply here at all, actually. If I understand correctly, "layout" has to do with turning spacing into braces and semicolons. A new line is a semicolon. A brace group is inserted around things that are aligned where a brace group actually makes sense. e.g. let x = y y = z in ... gets turned into roughly let { x = y ; y = z } in ... if-then-else however is just an expression, like a ternary operator, so it doesn't need any semicolons or braces. So layout is unrelated, except for the issue of `do` blocks inserting semicolons into `if-then-else` groups (and that's what DoAndIfThenElse fixes). -- Andrew PS. I am very glad someone is working on ghc-exactprint. It's a really important step in developing better Haskell tooling, imho. Doing that sort of thing right now with haskell-src-exts right now is a real pain (see half of the closed issues on hindent... about how it doesn't preserve formatting in many places.) On Tue, Mar 10, 2015 at 9:25 AM, Alan & Kim Zimmerman wrote:

Ok, adding a do does make a difference. More complexity.

Thanks Alan

On Tue, Mar 10, 2015 at 6:06 PM, Brandon Allbery wrote:

...
On Tue, Mar 10, 2015 at 12:01 PM, Alan & Kim Zimmerman < alan.zimm@gmail.com> wrote:

...
I am working on ghc-exactprint, and need to flag points where layout must be preserved, and considering the `if` statement.

My understanding of the layout rules for if then else is that the `then` has to start more to the right than the `if`.

I believe you are looking for DoAndIfThenElse.

-- brandon s allbery kf8nh sine nomine associates allbery.b@gmail.com ballbery@sinenomine.net unix, openafs, kerberos, infrastructure, xmonad http://sinenomine.net

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe

Reply

Sign in to reply online Use email software

Alan & Kim Zimmerman

2:14 p.m.

Thanks, I interpret layout as you say, but with the aim of preserving valid spacing if syntax elements are moved around as a result of AST modifications. My implementation flags those points, such as the contents of a let .. in, or a do expression. In terms of ghc-exactprint, I actually think the technique I am using of converting all the original absolute locations to relative ones, editing the AST and then outputting it should be transferrable to haskell-src-exts too. Alan On Tue, Mar 10, 2015 at 8:05 PM, Andrew Gibiansky < andrew.gibiansky@gmail.com> wrote:

Just to be clear, I *think* layout rules don't apply here at all, actually.

If I understand correctly, "layout" has to do with turning spacing into braces and semicolons. A new line is a semicolon. A brace group is inserted around things that are aligned where a brace group actually makes sense.

e.g.

let x = y y = z in ...

gets turned into roughly

let { x = y ; y = z } in ...

if-then-else however is just an expression, like a ternary operator, so it doesn't need any semicolons or braces. So layout is unrelated, except for the issue of `do` blocks inserting semicolons into `if-then-else` groups (and that's what DoAndIfThenElse fixes).

-- Andrew

PS. I am very glad someone is working on ghc-exactprint. It's a really important step in developing better Haskell tooling, imho. Doing that sort of thing right now with haskell-src-exts right now is a real pain (see half of the closed issues on hindent... about how it doesn't preserve formatting in many places.)

On Tue, Mar 10, 2015 at 9:25 AM, Alan & Kim Zimmerman
...
wrote:

...
Ok, adding a do does make a difference. More complexity.

Thanks Alan

On Tue, Mar 10, 2015 at 6:06 PM, Brandon Allbery wrote:

...
On Tue, Mar 10, 2015 at 12:01 PM, Alan & Kim Zimmerman < alan.zimm@gmail.com> wrote:

...
I am working on ghc-exactprint, and need to flag points where layout must be preserved, and considering the `if` statement.

My understanding of the layout rules for if then else is that the `then` has to start more to the right than the `if`.

I believe you are looking for DoAndIfThenElse.

-- brandon s allbery kf8nh sine nomine associates allbery.b@gmail.com ballbery@sinenomine.net unix, openafs, kerberos, infrastructure, xmonad http://sinenomine.net

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe

Reply

Sign in to reply online Use email software

Albert Y. C. Lai

12 Mar 12 Mar

1:01 p.m.

To torment you further, I and others have legal (Haskell 2010), bizzare examples at http://lpaste.net/81623 (Most of them were illegal in Haskell 98.)

Reply

Sign in to reply online Use email software

Alan & Kim Zimmerman

1:44 p.m.

Wow. Thanks, will put them in as test cases Alan On Thu, Mar 12, 2015 at 7:01 PM, Albert Y. C. Lai wrote:

To torment you further, I and others have legal (Haskell 2010), bizzare examples at

http://lpaste.net/81623

(Most of them were illegal in Haskell 98.)

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe

Reply

Sign in to reply online Use email software

3723

Age (days ago)

3725

Last active (days ago)

Download

7 comments

5 participants

tags

participants (5)

Alan & Kim Zimmerman
Albert Y. C. Lai
Andrew Gibiansky
Brandon Allbery
Vo Minh Thu