diff mbox series

[bug#33575] guix: lint: Add checker to check if inputs are sorted.

Message ID 20181202074210.31361-1-arunisaac@systemreboot.net
State Accepted
Headers show
Series [bug#33575] guix: lint: Add checker to check if inputs are sorted. | expand

Checks

Context Check Description
cbaines/applying patch success Successfully applied

Commit Message

Arun Isaac Dec. 2, 2018, 7:42 a.m. UTC 
* guix/scripts/lint.scm (check-inputs-should-be-sorted): New procedure.
(%checkers): Add it.
---
 guix/scripts/lint.scm | 21 +++++++++++++++++++++
 1 file changed, 21 insertions(+)

Comments

Ludovic Courtès Dec. 3, 2018, 1:31 p.m. UTC | #1 
Hello,

Arun Isaac <arunisaac@systemreboot.net> skribis:

> * guix/scripts/lint.scm (check-inputs-should-be-sorted): New procedure.
> (%checkers): Add it.
> ---
>  guix/scripts/lint.scm | 21 +++++++++++++++++++++
>  1 file changed, 21 insertions(+)
>
> diff --git a/guix/scripts/lint.scm b/guix/scripts/lint.scm
> index 2314f3b28..37e8a1ec5 100644
> --- a/guix/scripts/lint.scm
> +++ b/guix/scripts/lint.scm
> @@ -8,6 +8,7 @@
>  ;;; Copyright © 2017 Alex Kost <alezost@gmail.com>
>  ;;; Copyright © 2017 Tobias Geerinckx-Rice <me@tobias.gr>
>  ;;; Copyright © 2017 Efraim Flashner <efraim@flashner.co.il>
> +;;; Copyright © 2018 Arun Isaac <arunisaac@systemreboot.net>
>  ;;;
>  ;;; This file is part of GNU Guix.
>  ;;;
> @@ -301,6 +302,22 @@ of a package, and INPUT-NAMES, a list of package specifications such as
>                (package-input-intersection (package-direct-inputs package)
>                                            input-names))))
>  
> +(define (check-inputs-should-be-sorted package)
> +  ;; Emit a warning if inputs, native inputs or propagated inputs of PACKAGE
> +  ;; are not lexicographically ordered.

It’s something we rarely do so we’d get warnings for most packages.  As
a side effect, people may pay less attention to what ‘guix lint’ says.

As for the goal itself, I think sorting is a good idea when there are
lots of inputs (things like IceCat), but otherwise I personally don’t
think it matters that much.

What do people think?

Thanks,
Ludo’.
Tobias Geerinckx-Rice Dec. 4, 2018, 9:13 a.m. UTC | #2 
Ludo', Arun,

Ludovic Courtès wrote:
>> +  ;; Emit a warning if inputs, native inputs or propagated 
>> inputs
>> of PACKAGE
>> +  ;; are not lexicographically ordered.
>
> It's something we rarely do so we'd get warnings for most
> packages.  As
> a side effect, people may pay less attention to what ‘guix lint’
> says.

Even I agree :-)  There are valid reasons not to sort them.

> As for the goal itself, I think sorting is a good idea when
> there are
> lots of inputs (things like IceCat), but otherwise I personally
> don't
> think it matters that much.

Do we already check for duplication?

I sometimes order inputs for the same reason I sort module 
imports: to catch duplicates.  These are usually harmless and 
produce no errors.

Kind regards,

T G-R
Oleg Pykhalov Dec. 6, 2018, 12:42 a.m. UTC | #3 
Hello,

ludo@gnu.org (Ludovic Courtès) writes:

[…]

> It’s something we rarely do so we’d get warnings for most packages.  As
> a side effect, people may pay less attention to what ‘guix lint’ says.

I think this should not stop us from improving a linter and an option
like --misc-checks=sort-input,foo,bar could be used for such cases.

[…]

Oleg.
swedebugia Dec. 6, 2018, 12:31 p.m. UTC | #4 
On 2018-12-06 01:42, Oleg Pykhalov wrote:
> Hello,
> 
> ludo@gnu.org (Ludovic Courtès) writes:
> 
> […]
> 
>> It’s something we rarely do so we’d get warnings for most packages.  As
>> a side effect, people may pay less attention to what ‘guix lint’ says.
> 
> I think this should not stop us from improving a linter and an option
> like --misc-checks=sort-input,foo,bar could be used for such cases.

I agree with Oleg here. 

If many packages needs inputs to be sorted lets write a guix lint --sort
modelling the updater (that is make the changes in the work tree to be
committed).
Maxim Cournoyer Dec. 7, 2018, 1:08 p.m. UTC | #5 
swedebugia@riseup.net writes:

> On 2018-12-06 01:42, Oleg Pykhalov wrote:
>> Hello,
>> 
>> ludo@gnu.org (Ludovic Courtès) writes:
>> 
>> […]
>> 
>>> It’s something we rarely do so we’d get warnings for most packages.  As
>>> a side effect, people may pay less attention to what ‘guix lint’ says.
>> 
>> I think this should not stop us from improving a linter and an option
>> like --misc-checks=sort-input,foo,bar could be used for such cases.
>
> I agree with Oleg here. 
>
> If many packages needs inputs to be sorted lets write a guix lint --sort
> modelling the updater (that is make the changes in the work tree to be
> committed).

+1
宋文武 Dec. 8, 2018, 3:51 a.m. UTC | #6 
Arun Isaac <arunisaac@systemreboot.net> writes:

> * guix/scripts/lint.scm (check-inputs-should-be-sorted): New procedure.
> (%checkers): Add it.
> [...]
>  
> +(define (check-inputs-should-be-sorted package)
> +  ;; Emit a warning if inputs, native inputs or propagated inputs of PACKAGE
> +  ;; are not lexicographically ordered.

Hello, consider 'gspell', it has some native-inputs for build and some
for test:

    (native-inputs
     `(("glib" ,glib "bin")
       ("pkg-config" ,pkg-config)
       ("xmllint" ,libxml2)

       ;; For tests.
       ("aspell-dict-en" ,aspell-dict-en)
       ("xorg-server" ,xorg-server)))

Currently I'd seperated them by a comment like this.

If they are sorted, I have to add comment for each test input:

  `(("aspell-dict-en", aspecll-dict-en) ; for test
    ("glib" ,glib "bin")
    ("pkg-config" ,pkg-config)
    ("xmllint" ,libxml2)
    ("xorg-server" ,xorg-server))       ; for test

Which will be a little annoying...
swedebugia Dec. 8, 2018, 7:58 a.m. UTC | #7 
On 2018-12-08 04:51, iyzsong@member.fsf.org wrote:
> Arun Isaac <arunisaac@systemreboot.net> writes:
> 
>> * guix/scripts/lint.scm (check-inputs-should-be-sorted): New procedure.
>> (%checkers): Add it.
>> [...]
>>
>> +(define (check-inputs-should-be-sorted package)
>> +  ;; Emit a warning if inputs, native inputs or propagated inputs of PACKAGE
>> +  ;; are not lexicographically ordered.
> 
> Hello, consider 'gspell', it has some native-inputs for build and some
> for test:
> 
>     (native-inputs
>      `(("glib" ,glib "bin")
>        ("pkg-config" ,pkg-config)
>        ("xmllint" ,libxml2)
> 
>        ;; For tests.
>        ("aspell-dict-en" ,aspell-dict-en)
>        ("xorg-server" ,xorg-server)))
> 
> Currently I'd seperated them by a comment like this.
> 
> If they are sorted, I have to add comment for each test input:
> 
>   `(("aspell-dict-en", aspecll-dict-en) ; for test
>     ("glib" ,glib "bin")
>     ("pkg-config" ,pkg-config)
>     ("xmllint" ,libxml2)
>     ("xorg-server" ,xorg-server))       ; for test
> 
> Which will be a little annoying...

You convinced me sorting is a bad idea. Thanks for providing a good
argument :)
Ludovic Courtès Dec. 8, 2018, 1:29 p.m. UTC | #8 
Hello,

Tobias Geerinckx-Rice <somebody@not-sent-or-endorsed-by.tobias.gr>
skribis:

> Ludovic Courtès wrote:

[...]

>> As for the goal itself, I think sorting is a good idea when
>> there are
>> lots of inputs (things like IceCat), but otherwise I personally
>> don't
>> think it matters that much.
>
> Do we already check for duplication?

No but it would be a worthy check!

Ludo’.
Arun Isaac Dec. 8, 2018, 1:34 p.m. UTC | #9 
> If they are sorted, I have to add comment for each test input:
>
>   `(("aspell-dict-en", aspecll-dict-en) ; for test
>     ("glib" ,glib "bin")
>     ("pkg-config" ,pkg-config)
>     ("xmllint" ,libxml2)
>     ("xorg-server" ,xorg-server))       ; for test
>
> Which will be a little annoying...

I too find this convincing. It's not a good idea to enforce sorted
inputs all the time. If there is sufficient consensus, we can close this
bug report.
Maxim Cournoyer Dec. 9, 2018, 10:49 p.m. UTC | #10 
Hi,

Arun Isaac <arunisaac@systemreboot.net> writes:

>> If they are sorted, I have to add comment for each test input:
>>
>>   `(("aspell-dict-en", aspecll-dict-en) ; for test
>>     ("glib" ,glib "bin")
>>     ("pkg-config" ,pkg-config)
>>     ("xmllint" ,libxml2)
>>     ("xorg-server" ,xorg-server))       ; for test
>>
>> Which will be a little annoying...
>
> I too find this convincing. It's not a good idea to enforce sorted
> inputs all the time. If there is sufficient consensus, we can close this
> bug report.

Maybe our test inputs should have their own field? This would make their
raison d'être explicit and remove the need of using comments.

Maxim
宋文武 Dec. 10, 2018, 11:19 a.m. UTC | #11 
Arun Isaac <arunisaac@systemreboot.net> writes:

>> If they are sorted, I have to add comment for each test input:
>>
>>   `(("aspell-dict-en", aspecll-dict-en) ; for test
>>     ("glib" ,glib "bin")
>>     ("pkg-config" ,pkg-config)
>>     ("xmllint" ,libxml2)
>>     ("xorg-server" ,xorg-server))       ; for test
>>
>> Which will be a little annoying...
>
> I too find this convincing. It's not a good idea to enforce sorted
> inputs all the time. If there is sufficient consensus, we can close this
> bug report.

Yes, I think so.  Thank you!
宋文武 Dec. 10, 2018, 11:45 a.m. UTC | #12 
Maxim Cournoyer <maxim.cournoyer@gmail.com> writes:

> Hi,
>
> Arun Isaac <arunisaac@systemreboot.net> writes:
>
>>> If they are sorted, I have to add comment for each test input:
>>>
>>>   `(("aspell-dict-en", aspecll-dict-en) ; for test
>>>     ("glib" ,glib "bin")
>>>     ("pkg-config" ,pkg-config)
>>>     ("xmllint" ,libxml2)
>>>     ("xorg-server" ,xorg-server))       ; for test
>>>
>>> Which will be a little annoying...
>>
>> I too find this convincing. It's not a good idea to enforce sorted
>> inputs all the time. If there is sufficient consensus, we can close this
>> bug report.
>
> Maybe our test inputs should have their own field? This would make their
> raison d'être explicit and remove the need of using comments.

Yeah, something like:

  (package
    ...
    (inputs ...)
  (test:inputs ...)
  (test:native-inputs ...))

If we plan to support build packages with tests disabled, this would be
the way to go.  And due to how build works in guix, if tests are
disabled, it would be considered as a different derivation/package, so
the main use case may be:

  - I disable substitute servers to build all packages from sources
    locally.
  - I want to disable tests for some packages as they are too slow...

I don't have this use case now, and seperate package inputs will be a
big change, so I think the current way is totally ok.
Arun Isaac Dec. 18, 2018, 8:36 p.m. UTC | #13 
I'm closing this bug report since I believe we have reached a consensus
that this patch is unnecessary. Thank you all for your thoughts!
diff mbox series

Patch

diff --git a/guix/scripts/lint.scm b/guix/scripts/lint.scm
index 2314f3b28..37e8a1ec5 100644
--- a/guix/scripts/lint.scm
+++ b/guix/scripts/lint.scm
@@ -8,6 +8,7 @@ 
 ;;; Copyright © 2017 Alex Kost <alezost@gmail.com>
 ;;; Copyright © 2017 Tobias Geerinckx-Rice <me@tobias.gr>
 ;;; Copyright © 2017 Efraim Flashner <efraim@flashner.co.il>
+;;; Copyright © 2018 Arun Isaac <arunisaac@systemreboot.net>
 ;;;
 ;;; This file is part of GNU Guix.
 ;;;
@@ -301,6 +302,22 @@  of a package, and INPUT-NAMES, a list of package specifications such as
               (package-input-intersection (package-direct-inputs package)
                                           input-names))))
 
+(define (check-inputs-should-be-sorted package)
+  ;; Emit a warning if inputs, native inputs or propagated inputs of PACKAGE
+  ;; are not lexicographically ordered.
+  (define (check-inputs inputs-accessor input-type)
+    (unless (sorted? (map (match-lambda
+                            ((name input) name))
+                          (inputs-accessor package))
+                     string<?)
+      (emit-warning
+       package
+       (format #f (G_ "~a should be in lexicographic order") input-type))))
+
+  (check-inputs package-inputs (G_ "inputs"))
+  (check-inputs package-native-inputs (G_ "native inputs"))
+  (check-inputs package-propagated-inputs (G_ "propagated inputs")))
+
 (define (package-name-regexp package)
   "Return a regexp that matches PACKAGE's name as a word at the beginning of a
 line."
@@ -1032,6 +1049,10 @@  them for PACKAGE."
      (name        'inputs-should-not-be-input)
      (description "Identify inputs that shouldn't be inputs at all")
      (check       check-inputs-should-not-be-an-input-at-all))
+   (lint-checker
+     (name        'inputs-should-be-sorted)
+     (description "Verify that inputs are in lexicographic order")
+     (check       check-inputs-should-be-sorted))
    (lint-checker
      (name        'patch-file-names)
      (description "Validate file names and availability of patches")