Re: [GHC] #15509: `showEFloat` inconsistency introduced in base-4.12

12 Aug 2018

      #15509: `showEFloat` inconsistency introduced in base-4.12
-------------------------------------+-------------------------------------
        Reporter:  hvr               |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:  8.6.1
       Component:  Compiler          |              Version:  8.6.1-beta1
      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 hvr:

Old description:
...
While base-4.11 had a `showEFloat` that behaved like
{{{#!hs
showEFloat (Just n) | n <= 0 == showEFloat (Just 1)
}}}
For non-positive values, base-4.12 introduced an unfortunate
inconsistency:
{{{#!hs
showEFloat (Just n) | n < 0 == showEFloat (Just 1)
-- NB: *not* equivalent to     showEFloat (Just 0)
}}}
e.g. now we have
{{{#!hs
showEFloat (Just 2)    1.0 "" == "1.00e0"
showEFloat (Just 1)    1.0 "" == "1.0e0"
showEFloat (Just 0)    1.0 "" == "1e0"
showEFloat (Just (-1)) 1.0 "" == "1.0e0"
showEFloat (Just (-2)) 1.0 "" == "1.0e0"
}}}
iow, negative `n`s in the precision argument are now treated differently
from `(Just 0)` which is a source of subtle bugs for API consumers which
assume `showEFloat` to have "continuous total" semantics for the
precision argument.
New description:

 While base-4.11 had a `showEFloat` that behaved like

 {{{#!hs
 showEFloat (Just n) | n <= 0 == showEFloat (Just 1)
 }}}

 For non-positive values, base-4.12 introduced an unfortunate
 inconsistency:

 {{{#!hs
 showEFloat (Just n) | n < 0 == showEFloat (Just 1)
 -- NB: *not* equivalent to     showEFloat (Just 0)
 }}}

 e.g. now we have

 {{{#!hs
 showEFloat (Just 2)    1.0 "" == "1.00e0"
 showEFloat (Just 1)    1.0 "" == "1.0e0"
 showEFloat (Just 0)    1.0 "" == "1e0"
 showEFloat (Just (-1)) 1.0 "" == "1.0e0"
 showEFloat (Just (-2)) 1.0 "" == "1.0e0"
 }}}

 iow, negative `n`s in the precision argument are now treated differently
 from `(Just 0)` which is a source of subtle bugs for API consumers which
 assume `showEFloat` to have "continuous total" (or rather, saturated
 subtraction) semantics for the precision argument.

--

-- 
Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/15509#comment:2
GHC http://www.haskell.org/ghc/
The Glasgow Haskell Compiler