Smooth out the differences between `compiler/utils/Pretty.hs` and `libraries/pretty`

GHC has an internal copy of pretty in compiler/utils/Pretty.hs.

Todo:

• refactor GHC's copy to make it as similar as possible to pretty, without making any real code changes.
• apply the bug fixes that pretty received to GHC's copy, making sure not to pick up any possible new bugs in the process.

According to (1):

"There is one situation where the laws for pretty are ambiguous and leave room for choice. GHC decided one way and pretty the other."

• Find which law that is, and document the differences.

This hopefully allows us to close the following tickets for free: #1062 (closed), #1176 (closed) and #7666 (closed), maybe more.

Ideally we could remove GHC's copy altogether, but we're not there yet. GHC's copy uses FastString, which is supposedly needed for performance, whereas pretty uses String.

(1) https://github.com/haskell/pretty/issues/1

Trac metadata

Trac field	Value
Version	7.0.1
Type	Task
TypeOfFailure	OtherFailure
Priority	normal
Resolution	Unresolved
Component	Compiler
Test case
Differential revisions
BlockedBy
Related	#1062 (closed), #1176 (closed), #7666 (closed)
Blocking
CC
Operating system
Architecture

mentioned in issue #8809 (closed)

changed weight to 5

added Ttask Trac import labels

assigned to @trac-thomie

I have patches.

Replying to [ticket:10735#comment:103505 thomie]:

I have patches.

Lovely. Will we be seeing them on Phab soon?

On a related note, I've recently been thinking about error reporting with an eye for making the messages produced by GHC more useful to external tools. I've described one possible approach in ticket:8809#comment:101672 which would involve making Doc a sort of free functor, allowing the insertion of rich AST objects into the document. Monadic semantics would certain patterns of projecting out these annotations into vanilla documents quite convenient.

Unfortunately, the Hughes/Peyton-Jones pretty-printer isn't terribly well-suited for this as you need to know the width of a node's parent during construction. I've been meaning to explore adopting the Wadler/Leijen pretty-printer as Edward Kmett has already demonstrated that an implementation with annotations along the lines of my proposal is possible.

Ok, maybe you want to comment on https://github.com/haskell/pretty/issues/1#issuecomment-78297997, where Trevor Elliott said:

"I was hoping to extend the GHC pretty printer with the annotations, to get functionality like in the idris repl. Maybe extending their forked version with the annotations is the right path forward for that work :)"

Trac metadata

Trac field	Value
Related	#1062 (closed), #1176 (closed), #7666 (closed) → #1062 (closed), #1176 (closed), #7666 (closed), #8809 (closed)

Trac metadata

Trac field	Value
Differential revisions	- → D1130

mentioned in commit d7b053a2

mentioned in commit 53484d3d

mentioned in commit 25bc4062

mentioned in commit 9d24b060

mentioned in commit 85179b58

mentioned in commit f951ffc6

mentioned in commit 926e4288

mentioned in commit 2d1eae26

mentioned in commit 6f6d0821

Trac metadata

Trac field	Value
Differential revisions	D1130 → D1130, D1132

mentioned in commit 5d57087e

mentioned in commit 67576ddc

mentioned in commit bcfae08c

mentioned in commit 85bf76a8

There is 1 commit left for parity with pretty-1.1.2.0. I have not landed it yet, because I haven't been able to measure the supposed space/time improvement. The other 2 commits described below seem to suggest this commit doesn't actually work:

• "Pretty: improving the space/time performance of vcat, hsep, hcat"

There are 2 commits left for parity with pretty-1.1.2.1. I have not landed them yet, because they seem to cause more allocation in the compiler (see discussion in https://github.com/haskell/pretty/pull/9):

• "Resolve foldr-strictness stack overflow bug"
• "Special-case reduce for horiz/vert"

The commits are here: https://github.com/thomie/ghc/commits/pretty

mentioned in commit f903949b

changed milestone to %8.0.1

Milestone renamed

changed milestone to %8.2.1

It seems unlikely that this is going to happen for 8.0. Punting to 8.2.

closed

changed milestone to %8.0.1

This is pretty much done.

For parity with pretty-1.1.2.1 the following two commits would also be needed. I did not land them, because they seem to cause more allocation in the compiler (see discussion in ​https://github.com/haskell/pretty/pull/9):

• "Resolve foldr-strictness stack overflow bug"
• "Special-case reduce for horiz/vert"

The commits are here: ​https://github.com/thomie/ghc/commits/pretty

Trac metadata

Trac field	Value
Resolution	Unresolved → ResolvedFixed

reopened

Re-opening, because we'd really like to start using the real pretty library for GHC. It just needs someone to finish the work Thomas started!

Trac metadata

Trac field	Value
Resolution	ResolvedFixed → Unresolved

I've made some notes about this for my convenience: https://github.com/niteria/notes/blob/master/PrettyPrintingGHC.md#problems, it summarizes some issues around this.

Note that adinapoli has been doing some great work on this: https://github.com/haskell/pretty/pull/43.

added Pnormal label