[GHC] #14324: Consider deprecating STM invariant mechanism

[GHC]

#14324: Consider deprecating STM invariant mechanism
-------------------------------------+-------------------------------------
        Reporter:  bgamari           |                Owner:  (none)
            Type:  task              |               Status:  new
        Priority:  normal            |            Milestone:  8.6.1
       Component:  Compiler          |              Version:  8.2.1
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------
Description changed by bgamari:

Old description:

> fryguybob and I were recently discussing the STM invariant mechanism.
> This mechanism is, presumably, intended to expose problems from odd
> interleavings of transactions. However, fryguybob pointed out that the
> implementation currently takes so many locks that it very likely prevents
> these odd interleavings from occurring.
>
> In addition,
>  * the implementation doesn't handle nested STM invariants correctly
> (#7930)
>  * the locking behavior of the implementation was, until very recently,
> utterly wrong (#14310)
>  * the feature introduces quite a bit of complexity in the RTS
>  * the interface has essentially no users, as evidenced by a Hackage
> search and the fact that #14310 went unnoticed for years
>
> All of this raises the question: Is the STM invariants feature really
> where we want to spend our complexity budget? Perhaps it is time for this
> feature to quietly pass (after an appropriate deprecation period, of
> course).

New description:

 fryguybob and I were recently discussing the STM invariant mechanism (e.g.
 `Control.Monad.STM.check`). This mechanism is, presumably, intended to
 expose problems from odd interleavings of transactions. However, fryguybob
 pointed out that the implementation currently takes so many locks that it
 very likely prevents these odd interleavings from occurring.

 In addition,
  * the implementation doesn't handle nested STM invariants correctly
 (#7930)
  * the locking behavior of the implementation was, until very recently,
 utterly wrong (#14310)
  * the feature introduces quite a bit of complexity in the RTS
  * the interface has essentially no users, as evidenced by a Hackage
 search and the fact that #14310 went unnoticed for years

 All of this raises the question: Is the STM invariants feature really
 where we want to spend our complexity budget? Perhaps it is time for this
 feature to quietly pass (after an appropriate deprecation period, of
 course).

--

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14324#comment:1>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler