ar.exe: Bad file number
John Earle writes:
- Hardware. Operating System: 32-Bit Edition of Windows Vista with Service Pack 2 applied. CPU: Intel Pentium 4
- Software. MinGW 5.1.6 which is the current stable release. MinGW 5.1.6 does not install the latest version of gcc, however. It installs gcc 3.4.5.
- MSYS 1.0.11 which is the current version of MSYS. It installs GNU Make 3.81 among other things. The version of MSYS available at http://hackage.haskell.org/trac/ghc/wiki/Building/Preparation/Windows is 1.0.10.
- msysDTK-1.0.1 which is the current stable release and is the same version available at http://hackage.haskell.org/trac/ghc/wiki/Building/Preparation/Windows.
- Preexisting version of GHC is 6.10.4 which is the current stable release of GHC which was used to build GHC 6.10.4 from source.
- Those additional Haskell tools that were needed were downloaded from http://hackage.haskell.org/trac/ghc/wiki/Building/Preparation/Windows.
- !ActiveState Python 220.127.116.11 (current is 18.104.22.168, but is not appropriate).
A build script along with a build procedure was prepared. These results do not reflect what was needed to reach this point. Many hours of work were required to achieve these results. These results were not the results that were achieved initially. The directions provided for how to build GHC from source proved insufficient. A workaround which was not described is needed. The build script runs make, the fast test suite, and make install. The documentation is not built. In order for the test suite to complete Windows Firewall had to be turned off which was something that was also not specified in the directions. Where the test suite needed to be unpacked to or should be named was furthermore not specified. These issues among others make the build process very expensive. Once you have all your ducks in the row, it is easier going, but that is how it usually is in life. Before proceeding remove the xargs shell script from the path. The xargs shell script is the workaround described above.
Wait 1 hour and 1 minute.
make -C libraries all make: Entering directory `/Z/dev/ghc-6.10.4/src/libraries/ghc-prim' (echo dist/build/cbits/longlong.o `find dist/build -name "*_stub.o" -print`; find dist/build/GHC/Bool_split dist/build/GHC/Generics_split dist/build/GHC/Ordering_split dist/build/GHC/PrimopWrappers_split dist/build/GHC/IntWord32_split dist/build/GHC/IntWord64_split dist/build/GHC/Tuple_split dist/build/GHC/Types_split dist/build/GHC/Unit_split -name '*.o' -print) | xargs c:/MinGW/bin/ar.exe q dist/build/libHSghc-prim-0.1.0.0.a xargs: c:/MinGW/bin/ar.exe: Bad file number make: *** [dist/build/libHSghc-prim-0.1.0.0.a] Error 126
In prior attempts I did not receive this error so early. Notice that the version of ar is c:/MinGW/bin/ar.exe. In http://hackage.haskell.org/trac/ghc/wiki/Building/Preparation/Windows c:/MinGW/bin/ occurs in the path before the preexisting instance of GHC just as is the case here. On a previous attempts I have gotten "xargs: c:/ghc/ghc-6.10.4/bin/ar.exe: Argument list too long" and "xargs: c:/ghc/ghc-6.10.4/bin/ar.exe: Bad file number". The path to ar appeared hard coded which is confusing. I moved the xargs shell script so that it appeared front most in the path once more and resumed the build process from where it left off. Wait 3 hours and 5 minutes. The build process completed successfully. At 2 hours and 42 minutes later after the build process was resumed an empty null file is created in the "Z:\dev" parent directory which an immediate child of the root directory. Regardless of how deeply nested the project directory is, it always appears in the dev directory. If all goes well the build process has consistently taken 4 hours and 6 minutes. The day before the build was completed with the workaround applied prior to carrying out the procedure. Today, when the workaround was applied only once an error was encountered and this resulted in additional unexpected test suite failure, namely: conc049(normal). These were yesterdays results:
OVERALL SUMMARY for test run started at The current date is: 2009.10.22 Enter the new date: (yy-mm-dd) 2403 total tests, which gave rise to 12815 test cases, of which 0 caused framework failures 10766 were skipped 1955 expected passes 76 expected failures 1 unexpected passes 17 unexpected failures Unexpected passes: system001(normal) Unexpected failures: 2816(ghci) DoParamM(normal) cabal01(normal) drvfail006(normal) drvfail008(normal) getPermissions001(normal,normal) ghci028(ghci) mod133(normal) net001(normal) signals004(normal) tc183(normal) tc217(normal) tc220(normal) tc223(normal) tc232(normal) tcfail126(normal)