Add exact print proposal by jappeace · Pull Request #5 · haskell/cabal-proposals

jappeace · 2025-12-23T14:05:49Z

No description provided.

ulysses4ever · 2025-12-23T14:22:16Z

@jappeace could you explain what's the relation to the earlier HF proposal?

Mikolaj · 2025-12-23T15:48:45Z

@jappeace could you explain what's the relation to the earlier HF proposal?

The start of the document says "Note that this is a copy of the TWG proposal: haskellfoundation/tech-proposals#6", but indeed I'm curious how literally this is a copy, in particular, if the experience gained when implementing and discussing lead to some modifications or maybe if this is the baseline over which the modifications are expected to be made.

ulysses4ever · 2025-12-23T15:56:13Z

There's probably no harm in copying old stuff, but I personally hoped that we can make some profit out of our process and discuss some details that poped up during the past couple months in a new document. If creating a whole new document sounds like too much, a new section in the copy would be fine as well...

jappeace · 2025-12-24T08:04:00Z

I put the somewhat relevant parts of TWG proposal into the relevant sections of the cabal proposal template.
and filled out some other headings, especially open questions is a fair bit different now.

unlike the original proposal says, the common stanzas have been investigated here:
haskell/cabal#11277

and unlike the original proposal says, comment parsing is now pretty solid with this PR:
haskell/cabal#11252
(my zurich hack prototype did only one or two locations, leana made it work for basically any valid position).

I think these are the only two major implementation issues left from the original TWG proposal:

+ Add support for comma printing.
+ add support for conditional branches.
  See the technical content section for more details on this.

Both of these are investigated by a prototype focussing on "trivia", because FieldGrammar handles both conditionals as well as the comma's.
haskell/cabal#11227 (comment)

ffaf1 · 2026-01-29T18:52:25Z

Hello, we discussed this proposal tonight ad the cabal dev-meeting.

We are generally happy with it, but there are things we would like to get a better idea of.

Interested parties. There are a lot of projects using Cabal (the library). It is nice to see “beneficiaries” listed, but the ideal situation would be to investigate who are the major users of Cabal, check who could be affected, collect their feedback before this proposal is accepted. Such thorough changes are bound to cause code breakage! Which leads to
Backwards Compatibility / Migration. The section is quite light on details, we need to know more, especially whether there is a simple path to upgrade and whether there are things which could require a more expensive workaround.
Impact on maintainers. We are here for the long run, so a sober assessment of how much more work to — e.g. — add to the spec, etc. would be interesting to know.

Thanks for having written this document!

geekosaur · 2026-01-29T18:59:15Z

I'd like to note that the current POC is a step in the direction of answering some of these questions (I did mention this in the meeting), in particular testing it against clc-stackage should give an idea of both who it affects and what kinds of mitigations and migration strategies will be required.

Mikolaj · 2026-01-29T19:15:58Z

Right, the POC is a great contribution. Maybe it could serve to update the proposal a bit, especially the parts the maintainers are keen about, as listed by Francesco? This proposal, as opposed to the TWG one, has the benefit of being submitted after some considerable prototyping and some feedback, so it can stand on much less hypothetical foundations than the TWG one. It should now be easier to involve many cabal "stakeholders" in the discussion, soliciting feedback about, e.g., backward compatibility and potential migration policy.

In any case, thank you very much for all this recent work, we are thrilled to see such a professional approach and we hope this can be brought to fruition with minimal friction.

jappeace · 2026-02-07T17:41:10Z

@ffaf1 Please help me understand what defines a major user, Cabal has around 250 direct depenencies on hackage.
I presume we're not intending to elicit feedback from all of these?

The other points I should be able to address with current information.

ffaf1 · 2026-02-12T11:56:29Z

@jappeace Apologies for the late reply, I am not ignoring you, I am just trying to check which project using Cabal actually e.g. goes through GPD datatype instead of “just” using functions which are opaque with regard to GPD.

It would of course be silly and wrong to ask you to collect feedback from each Cabal dependency.

I recall in the last cabal-dev meet @geekosaur had an example of such a “probably impacted” dependency, but I forgot which one it was.

geekosaur · 2026-02-12T16:37:55Z

I think I mentioned the custom Setup of gtk2hs and gi-gtk as things which I recall poking at GPD internals. I'm also pretty sure I've seen others, all involving custom Setup.

jappeace · 2026-02-21T21:34:54Z

I cleaned up the document, please review whenever you've time 🙂

Mikolaj · 2026-02-21T22:01:17Z

Oh, great! Yes, we will.

Mikolaj · 2026-03-12T19:06:38Z

People expressed the need for a few more days to review this proposal. I'm sure we will decide on it in the meeting in 2 weeks. @jappeace: please let us know whenever there's any deadline either regarding this proposal or the implementation PRs themselves --- almost all the work on cabal is done by volunteers, so we can't commit to any review timelines, but we do try our best, especially if notified in advance. Thank you!

ffaf1

Well documented, clear, thorough document, great job!
Some comments and questions for you.

jappeace · 2026-03-15T15:07:23Z

@Mikolaj I understand this is mostly driven by volunteers yes, thanks for your time :)

Mikolaj · 2026-03-26T18:07:25Z

@jappeace asked to postpone approving the proposal, so the next deadline it the cabal chat in 2 weeks.

ulysses4ever · 2026-04-09T15:17:20Z

Would it be possible to showcase the approach that you take by providing examples of how some of the common exact-print-related tasks could be solved after you're done? For instance, how to implement (some of) cabal-add or how to implement the addition of an exposed module.

Mikolaj · 2026-04-09T18:02:25Z

Regarding the procedural side, since there are many outstanding comments right now, we have decided once more to postpone the decision about this proposal by two weeks. Apologies!

ulysses4ever · 2026-04-10T23:52:58Z

+We developed an alternative approach in the barbies section.
+
+### (TWG) Namespace approach
+where `hackagePackage` is a cabal package found on Hackage.


Another half-sentence?

ulysses4ever · 2026-04-11T00:34:02Z

+In this approach we try to eat the cake and have it too with some clever type machinery.
+We want to be able to annotate `GenericPackageDescription` with exact data,
+however we also want to keep it backwards compatible.
+Here we we're inspired by the [barbies](https://hackage.haskell.org/package/barbies) 


Suggested change

Here we we're inspired by the [barbies](https://hackage.haskell.org/package/barbies)

Here we are inspired by the [barbies](https://hackage.haskell.org/package/barbies)

ulysses4ever · 2026-04-11T00:39:07Z

+This will allow us to retain the shape of the parse result. 
+Solving the following problems:
+
+In field grammar we sidestep the problem where we fail to lookup trivia.


What problem is it, can you elaborate? The whole paragraph is impossible to follow for me. I'd need to get more elaborated discussion.

It's just hard to debug because you don't know if:

The key is actually missing

the path got mangled somehow.

One of the above typeclasses in field grammar didn't pass the argument

If this is the case you need to track down which. To do this you need to write golden tests and dum p the AST.

It's just time consuming, and it appears kidn of fragile.
This barbies approach is compiler driven, which is a lot of work but it's a lot less error prone

ulysses4ever · 2026-04-11T00:47:40Z

In one of the earlier attempts at it from you Bodigrim asked a bunch of questions about handling particular syntax. E. g. a comma in a list of packages inside build-depends (for instance) can go on either line. Will you be able to capture this sort of differences with your approach? Generally, there were a bunch of good questions in the PRs you submitted earlier. It'd be wise to go over then and address them in the proposal. E. g. one of the big points was whether you have to touch GPD or would it be possible to stick with Field. This is only mentioned in passing in the proposal, while I'd expect a more in-depth discussion of this decision.

Bodigrim · 2026-04-20T22:50:41Z

+All the difficulty lies in figuring out where to place a dependency;
+we made a decision here to do it in the main library assuming it exists.
+We also assumed there would be no conditionals.


Anecdotally, one of the very first feature requests for cabal-add was "Can I add a dependency to a common stanza?" How would your approach handle it? Common stanzas are not even present in GPD.

I suppose you'd use the field gpdCommonStanza, see updated proposal

Bodigrim · 2026-04-20T22:51:51Z

+      let lib = condTreeData tree
+       in tree { condTreeData = lib { exposedModules = modName : exposedModules lib } }
+```
+This is the kind of operation that `cabal-add` or HLS could use.


This is the kind of operation that cabal-add already supports and HLS uses (through cabal-add).

addressed this

Bodigrim · 2026-04-20T22:52:17Z

+to modify it's own files?
+It's hard to guarantee stability across projects if many functionalities
+are distributed across many projects.
+For example a newly introduced tool called [cabal-add](https://github.com/Bodigrim/cabal-add), would need to take into account any 


As of 2026, it's no longer "a newly introduced" ;)

haha, age old text 😅

Bodigrim · 2026-04-20T22:54:32Z

+The byte-for-byte roundtrip property holds where `hackagePackage` is a cabal package found on Hackage.
+
+We explored two approaches.
+The namespace (trivia-tree) approach works but progress was slow because:


A forward reference to a section, explaining the term would be helpful. At this point a reader has no idea what it is about.

it's now explained inline

Bodigrim · 2026-04-20T23:05:11Z

+type GenericPackageDescription    = GenericPackageDescriptionWith HasNoAnn
+type GenericPackageDescriptionAnn = GenericPackageDescriptionWith HasAnn
+
+data GenericPackageDescriptionWith (m :: HasAnnotation) = GenericPackageDescription


By the time we arrive to GPD, common stanzas have been already inlined and (I think) normalised. So any positions and annotations can refer to places far away from. Say, a list of library dependencies will contain items from any stanza or from the library section itself. How would you reconstruct the original Cabal file? How would you insert a new dependency in such list?

It's deffered, now explained more elaborately in the proposal but we've a PR for that open too (although leana wants to work more on it before opening it to review).

Bodigrim · 2026-04-20T23:14:13Z

+This keeps the machinery simple and avoids the pattern-match exhaustiveness problems
+that Trees That Grow can introduce in GHC's own codebase.


I don't follow: the whole point of TTG is to avoid the issue with non-exhaustive pattern-matching?..

Yes this was misdescribed, we just have a relatively simple HasAnn | HasNoAnn, TTG is more complex but addresses that as well in their own way.

Bodigrim · 2026-04-20T23:14:47Z

+  Our closed type family with two constructors avoids this.
+
+ **cabal-fields / Field-level manipulation** (raised by @mpickering referencing cabal-add):
+  working at the `Field` level avoids touching GPD but loses type safety.


I don't really follow, what type safety are we talking about?

you no longer have to work with raw strings, the text reflects that.

Bodigrim · 2026-04-20T23:17:42Z

+A first simple roundtrip test based on the barbies approach is expected soon.
+
+### Exact printing
+The `pretty` library doesn't have a newline primitive, and I find it hard to position elements exactly.


Hmmm? Why did we suddenly jump to pretty library? What does it have to do with the rest of the proposal?

we found out that pretty isn't going to work in the trivia tree approach. This is why we introduced exact pretty which seems to make things a lot easier.

Bodigrim · 2026-04-20T23:20:42Z

+but code that is polymorphic over the GPD parameter or that
+pattern-matches on `GenericPackageDescriptionWith` directly will see `LibraryWith m`.
+
+Additionally, common stanza merging will be postponed to allow


Postponing merging of common stanzas is a significant design change, which I have not seen discussed above or agreed upon.

There is an entire seciton on this now

Bodigrim · 2026-04-20T23:22:09Z

+  e.g. `cabal init` could be described via `GenericPackageDescription`.
+
+The HLS project would benefit from this effort;
+it could add a dependencies plugin for example: https://github.com/haskell/haskell-language-server/issues/155


The referenced issue is closed as completed.

haha partially yes.

Bodigrim · 2026-04-22T20:35:39Z

+([#4617](https://github.com/haskell/haskell-language-server/pull/4617)).
+
+However, several more fine-grained operations remain open
+that are difficult to implement at the `Field` level:


This phrasing (perhaps inadverently) seems to suggest that the operations listed below are not available through cabal-add. That's not so: both renaming and adding exposed-modules / other-modules are supported in cabal-add, even if HLS plumbing remains unfinished.

how are you even supposed to know that? neither the readme nor the help command tells you what options are available 😅

cabal-add-the-library, not cabal-add-the-executable. See https://hackage-content.haskell.org/package/cabal-add-0.2/docs/Distribution-Client-Add.html#t:AddConfig and https://hackage-content.haskell.org/package/cabal-add-0.2/docs/Distribution-Client-Rename.html#t:RenameConfig

very good, thanks 🙏🏽

Bodigrim · 2026-04-22T20:37:42Z

+ `Nest` distributes over `Concat` and `StickyConcat`
+  but does **not** override `Place`:
+
+```hs


hs -> haskell

very fancy

This reverts commit 81a9472.

Mikolaj · 2026-04-23T18:31:38Z

We've accepted the proposal at today's regular cabal developer's meeting. This should not stop the discussion either here or in the PRs that will implement the proposal, because the design will naturally evolve all the time. Thank you everybody!

ulysses4ever · 2026-08-13T17:36:25Z

Since the proposal was accepted, should we merge, perhaps, @Mikolaj?

Mikolaj · 2026-08-13T19:51:53Z

Huh, that's true, I missed that. But now we have a new, even better proposal for the same feature, so maybe it's less noise if we don't keep both merged? @jappeace: how do you prefer to settle this?

jappeace · 2026-08-14T08:36:34Z

better we focus on that other one #7

Add exact print proposal

d5d75b4

jappeace mentioned this pull request Dec 23, 2025

Cabal exact printer in stages? (and other grand design questions) haskell/cabal#11227

Open

add links explaining where this comes from

c2fbb03

Finish sentence

d0886e8

jappeace and others added 4 commits February 12, 2026 18:54

Add current updates

223614c

document choices regarding exact printer

4b630ff

fix typos

df06ff6

Clean up proposal

e3a2bb2

Fix spell errors

3b8233a

ffaf1 reviewed Mar 13, 2026

View reviewed changes

jappeace added 3 commits March 15, 2026 17:00

Add barbies alternative to implementing exact print

b01f2b9

encodes just the namespace

12e0ff4

point out backwards compatibility

165d75a

jappeace added 2 commits March 28, 2026 22:00

Remove the stuff we're not going to do

2678a73

We decided on pretty already

5b37178

ulysses4ever reviewed Apr 10, 2026

View reviewed changes

ulysses4ever reviewed Apr 11, 2026

View reviewed changes

jappeace-sloth mentioned this pull request Apr 15, 2026

Address substantive review comments on exact print proposal jappeace/cabal-proposals#1

Closed

3 tasks

Address most review comments

f09a654

Bodigrim reviewed Apr 20, 2026

View reviewed changes

jappeace added 7 commits April 21, 2026 21:19

Address bodigrim's feedback

031786a

Repair where trivia is stored

7c266df

Add note on how to address the spaces betweeon colons

226feb9

address better TTG implemenetation

b3de179

Address leana feedback

4148208

Remove depracated open questions

08e5772

Do another test round to see if we got breakage

62683b5

Bodigrim reviewed Apr 22, 2026

View reviewed changes

jappeace added 4 commits April 22, 2026 23:07

Add note on cabal add support

2590a1a

fix hs to haskell

b7185fd

Add note on fancyAddAlgorithm.

81a9472

very fancy

Revert "Add note on fancyAddAlgorithm."

4723d0e

This reverts commit 81a9472.

ffaf1 approved these changes Apr 23, 2026

View reviewed changes

Mikolaj approved these changes Apr 23, 2026

View reviewed changes

leana8959 mentioned this pull request Jul 27, 2026

Cabal Exactprint new proposal #7

Open

jappeace closed this Aug 14, 2026

	Here we we're inspired by the [barbies](https://hackage.haskell.org/package/barbies)
	Here we are inspired by the [barbies](https://hackage.haskell.org/package/barbies)

		This keeps the machinery simple and avoids the pattern-match exhaustiveness problems
		that Trees That Grow can introduce in GHC's own codebase.

Uh oh!

Conversation

jappeace commented Dec 23, 2025

Uh oh!

ulysses4ever commented Dec 23, 2025

Uh oh!

Mikolaj commented Dec 23, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ulysses4ever commented Dec 23, 2025

Uh oh!

jappeace commented Dec 24, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ffaf1 commented Jan 29, 2026

Uh oh!

geekosaur commented Jan 29, 2026

Uh oh!

Mikolaj commented Jan 29, 2026

Uh oh!

jappeace commented Feb 7, 2026

Uh oh!

ffaf1 commented Feb 12, 2026

Uh oh!

geekosaur commented Feb 12, 2026

Uh oh!

jappeace commented Feb 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Mikolaj commented Feb 21, 2026

Uh oh!

Mikolaj commented Mar 12, 2026

Uh oh!

ffaf1 left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jappeace commented Mar 15, 2026

Uh oh!

Mikolaj commented Mar 26, 2026

Uh oh!

ulysses4ever commented Apr 9, 2026

Uh oh!

Mikolaj commented Apr 9, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ulysses4ever commented Apr 11, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Mikolaj commented Dec 23, 2025 •

edited

Loading

jappeace commented Dec 24, 2025 •

edited

Loading

jappeace commented Feb 21, 2026 •

edited

Loading

ffaf1 left a comment •

edited

Loading