[GHC] #11744: Latest Xcode update violates POSIX compliance of `nm -P`

[GHC]

#11744: Latest Xcode update violates POSIX compliance of `nm -P`
---------------------------------+----------------------------------------
        Reporter:  hvr           |                Owner:
            Type:  bug           |               Status:  new
        Priority:  normal        |            Milestone:
       Component:  Build System  |              Version:
      Resolution:                |             Keywords:
Operating System:  MacOS X       |         Architecture:  Unknown/Multiple
 Type of failure:  None/Unknown  |            Test Case:
      Blocked By:                |             Blocking:
 Related Tickets:                |  Differential Rev(s):
       Wiki Page:                |
---------------------------------+----------------------------------------
Description changed by hvr:

@@ -80,0 +80,13 @@
+
+ ----
+
+ Alternative workaround (since it turns out that Apple's `nm` violates
+ POSIX even more than assumed by not even supporting `nm -P -t x`):
+
+ Have `deriveConstants` derive a known value, e.g. the value `0x1234` via
+ `char derivedConstantCONTROL_GROUP[1+0x1234];`, and if the value for
+ `CONTROL_GROUP` comes back as `18016` we know that we need to fixup all
+ decimal-wrongly-as-hex parsed integers, if it comes back as `4660` we
+ parsed correctly; any different value is a non-recoverable error
+
+ This would also  serve as a general integrity check for //all// platforms

New description:

 TLDR: Apple messed up again and broke POSIX compliance


 luite ran into problems when he installed the latest Xcode update because
 the derived platform constants were off.

 He provided me the output of `nm -P ./includes/dist-
 derivedconstants/header/tmp.o` which turned out to have numbers which
 didn't make sense:

 {{{
 _derivedConstantAP_STACK_SPLIM C 1025 0
 _derivedConstantBITMAP_BITS_SHIFT C 7 0
 _derivedConstantBLOCKS_PER_MBLOCK C 253 0
 _derivedConstantBLOCK_SIZE C 4097 0
 _derivedConstantCINT_SIZE C 5 0
 _derivedConstantCLONG_LONG_SIZE C 9 0
 _derivedConstantCLONG_SIZE C 9 0
 _derivedConstantDOUBLE_SIZE C 9 0
 _derivedConstantDYNAMIC_BY_DEFAULT C 2 0
 ...
 }}}

 Which with an older Xcode version would have looked like

 {{{
 _derivedConstantAP_STACK_SPLIM C 401 0
 _derivedConstantBITMAP_BITS_SHIFT C 7 0
 _derivedConstantBLOCKS_PER_MBLOCK C fd 0
 _derivedConstantBLOCK_SIZE C 1001 0
 _derivedConstantCINT_SIZE C 5 0
 _derivedConstantCLONG_LONG_SIZE C 9 0
 _derivedConstantCLONG_SIZE C 9 0
 _derivedConstantDOUBLE_SIZE C 9 0
 _derivedConstantDYNAMIC_BY_DEFAULT C 2 0
 ...
 }}}

 So obviously `nm -P`'s output changed from hexadecimal to (non-compliant)
 decimal formatting of numbers.

 For reference, here's how POSIX specifies the semantics of `nm`:

 http://pubs.opengroup.org/onlinepubs/9699919799/utilities/nm.html

 ----

 > `-P`
 >    Write information in a portable output format, as specified in the
 STDOUT section.

 ...

 > If the `-P` option is specified, the previous information shall be
 displayed using the following portable format. The three versions differ
 depending on whether `-t d`, `-t o`, or `-t x` was specified,
 respectively:
 >
 > {{{"%s%s %s %d %d\n", <library/object name>, <name>, <type>, <value>,
 <size>}}}
 >
 > {{{"%s%s %s %o %o\n", <library/object name>, <name>, <type>, <value>,
 <size>}}}
 >
 > {{{"%s%s %s %x %x\n", <library/object name>, <name>, <type>, <value>,
 <size>}}}

 ...

 > '''If `-P` is specified, but `-t` is not, the format shall be as if `-t
 x` had been specified.'''

 ----


 In the event that Apple can't be bothered to fix this bug (I've had bad
 experience with Apple and standard-compliance in the past, hence my
 pessimism) we could try to workaround by using `nm -P -t x` when building
 on OSX...

 ----

 Alternative workaround (since it turns out that Apple's `nm` violates
 POSIX even more than assumed by not even supporting `nm -P -t x`):

 Have `deriveConstants` derive a known value, e.g. the value `0x1234` via
 `char derivedConstantCONTROL_GROUP[1+0x1234];`, and if the value for
 `CONTROL_GROUP` comes back as `18016` we know that we need to fixup all
 decimal-wrongly-as-hex parsed integers, if it comes back as `4660` we
 parsed correctly; any different value is a non-recoverable error

 This would also  serve as a general integrity check for //all// platforms

--

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/11744#comment:4>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler