Bounded deepseq, including support for generic deriving

Latest on Hackage:

This package is not currently in any snapshots. If you're interested in using it, we recommend adding it to Stackage Nightly. Doing so will make builds more reliable, and allow to host generated Haddocks.

BSD3 licensed by Andrew G. Seniuk
Maintained by Andrew Seniuk

For additional documentation, browse to ./HTML


** ->
- this is API-breaking because the semantics of forcep have changed
- (see [Note] below these, earlier explanations)
- IsString instance (-XOverloadedStrings)
- how could I not have thought of this before?
> ghci
>> import Control.DeepSeq.Bounded
>> let x = ( 'a' , ( undefined::Bool , 0xC ) ) :: (Char,(Bool,Int))
- before:
>> rnfp (compilePat ".(..)") x
>> rnfp (compilePat ".(!.)") x
*** Exception: Prelude.undefined
- now (but the above still works of coure):
>> rnfp ".(..)" x
>> rnfp ".(!.)" x
*** Exception: Prelude.undefined
- required no code change, just addition of this instance:
instance IsString Pattern where fromString = compilePat
- hahaha!!
- I guess this is something you'd definitely want to do with
any DSL having a concrete syntax
- (not as a necessary consequence of OverloadedStrings, but
as an obvious type change which they make practical)
- the new semantics are neither a superset nor a subset of the previous
- also, forcep_ is a little stronger
- domain of definition is a superset of the previous
- and in fact forcep and forcep_ are now synonymous (i.e. forcep_ = forcep)
- the underscore version is deprecated for removal in 0.9
- this applies to deepseqp/deepseqp_ likewise
- the specific change to semantics is:
- forcep used to accept only String first argument (literal or variable)
- but now it accepts Pattern first argument, or String literal
- and forcep_ used to accept only Pattern first arg.
- but now accepts that, or String literal
- the only function in the new API still accepting a String variable
argument (representing a pattern) is compilePat (hopefully)
- that has a tidiness about it ->
- still trying to decide on policy regarding cpphs
- this is all motivated by one pesky rogue anonymous build spammer
- had a spot of trouble over this, but learned a few tricks
- hopefully this upload is the ticket; sorry for the extra version uploads ->
- documentation touch-ups, mostly
- truncatePat did actually have an API-breaking change here (sorry!
at least it had been announced as coming in 0.7 in the comment

** ->
- although slated for early March, I'm just going ahead
and releasing 0.7 now, which finalises the grammar, and
largely closes the book on the first volume of this enterprise
- plans from here to version 1.0:
- definitely get rid of the T* nodes, and just rely on the attribute
- still undecided to refactor PatNode as product instead of sum...
- if do refactor, it'll be easier with T* nodes absorbed
- if I can get better test suites in there, that would be swell
- if things pass thorough testing, that will bring me to 0.8
- in case a major bump itch was pending for any reason,
NOW's the time to pull the bumper, rather than later!
- between 0.8 and 1.0, there needs to be work on Seqable
- though maybe there's not much to do, since it's (A) so minimalist,
and (B) designed for use by auto-instrumentation tools
- so basically, if everything in deepseq-bounded has polish
and is well-tested, there's nothing holding back 1.0 I think...
- just go 0.9 then give it a grace period, like let's say
two months (seeing 0.9.1 0.9.2 ...), and after that,
when things feel right, upload whatever I've got as 1.0 ->
- everything relating to attoparsec has been ripped out
- quite a bit lighter on the dependencies as a result ->
- now have an H98 parser for the new grammar, and it's quite
a bit shorter than the attoparsec
- I will yank out the attoparsec parser completely in 0.6.2!
- non-manual USE_CPPHS flag added (also in seqaid and in leaky)
so that client build system can decide whether cpp is locally available
or whether they need to install cpphs (and, if they've configured their
path correctly, to use it)
- flanking whitespace within type constraints lists is now
accepted (ignored)
- whitespace is not preserved beyond the parser; the need would
have to be demonstrated to support that
- yet further decrufting of Compile_new_grammar.hs ->
- realised I should be using Manual: True in all my .cabal flags,
or else the constraint solver is free to toggle them!! ->
- VACANT_HASH flag removed (permanent value = False)
- further decrufting of Compile_new_grammar.hs
- HASKELL98_FRAGMENT tested a bit more (compiles at least,
and the tests run, but there's still no H98 parser...)
- manual NFData instance of Pattern (instead of generic deriving)
so can use it to force non-interleaving of output with warnings,
even in H98 (and has other uses when debugging)
- a bit of sanitising of .cabal flag descriptions, and of NFDataP.handleAttrs
- a few incidental bug fixes; and a few recognised but not yet fixed... ->
- bugfixes so can build with NEW_IMPROVED_PATTERN_GRAMMAR set False
- got rid of three transient flags, which are now "permanently defaulted"
to the values shown:
(To read about these, consult the hackage page which documents
these flags; all of them are departures from the new grammar as published.)
- decrufting of Compile_new_grammar.hs

** 0.5.5 ->
- the transition-5-6-7.html file contains additional details.
- it is in the ./HTML directory of the source distribution, or online at
- many misc. bug fixes
- much polishing of documentation (corrections, elaborations, refinements)
- renamed module "...Bounded.Generics" to "...Bounded.Generic"
- changed the pattern concrete syntax slightly (now more compact)
- refer to
- more than "slightly", in the end!
- added nice data family FF :)
- changed PatNode so all constructors take a single parameter of
(new) type PatNodeAttrs
- several new capabilities (refer to the API docs for PatNodeAttrs)
- simplified the PatNode type itself
- using attoparsec to build the pattern DSL parser
- PatAlg.hs -> PatUtil.hs
- isSubPatOf -> subPat
- I hesitated to shed the "is", but the Bool result value
pretty much clears up any possible ambiguity there
- the ambiguity (which still remains) is, which is the
subpattern, which the host pattern, as the types don't
help you there!
- but we're pretty safe, corresponds to infix use as
if it were the mathematical inclusion symbol
- type constraints are now handled as a prefix modifier like
any other (except that internally they still have their
own PatNode's TI, TR etc.)
- more...

comments powered byDisqus