Building OpenJDK9 on MSYS2

Thu Oct 5 08:10:50 UTC 2017

Hello Peter,

On 2017-10-04 21:15, Peter Budai wrote:
> Hi Magnus,
>
> Thanks for the quick reply I’ll check these patches with msys2.
>
> Let me specify with more details what I’d like to achieve: I’d like to build OpenJDK9 with MSYS2 MINGW64 environment using gcc toolchain. (I’m not sure how familiar are you with MSYS2, but there are 3 different environments: MSYS2, MINGW32 and MINGW64). In theory MINGW64 with gcc is the closes you can get on Windows platform as a gcc unix like build environment, which produces still a native 64-bit executable on Windows.
>
> I’m not very familiar with OpenJDK yet, so therefore I’d like to hear your opinion: how realistic is that?
Sorry to disappoint, but I would say that requires major work. There is 
a strong historic assumption that windows builds are done using Visual 
Studio. We have abstracted away some of it in configure (see 
TOOLCHAIN_TYPE), but it's very far from enough to change compiler 
environment for a Windows build. The native sources are also bound to 
make a lot of such assumptions. I would expect the changes needed to be 
in the thousands of lines of code.

When we say supporting the build in msys2 instead of cygwin, we just 
mean using msys2 as the unix emulating layer for our tools like 
make/bash/grep/sed etc.

One think I have done successfully is running the build in WSL (Windows 
Subsystem for Linux), but that isn't all that helpful as WSL for 
practical purposes is more or less like running Linux in a VM, so the 
build sees a Linux system and builds a Linux binary.
> As a side note: with MINGW64 I have managed to run configure phase successfully for OpenJDK. The compile process has also started and went for a while, but interestingly I run into some kind of race conditions as make stopped with an error. Using LOG=debug I have fond the failing line and then copying the failed command and pasting it to the bash prompt it successfully generated the output target, and then the build process run further when a similar situation happened. Also pasting the failed command run in the bash without any problem, and build continued… until the next.
Without seeing the errors I can't say much. I very much doubt that you 
are running with gcc as the compiler though. Configure isn't easily 
fooled into using a different compiler to what it prefers, and I would 
expect things to crash and burn pretty early if you actually did.

/Erik
>
> I have tried to run make JOBS=1, but did not help, strangely I have still seen in the log make[3] and make[4] logs which suggested that there are more than one make jobs were running. Also tried .configure --with-output-sync=recurse without success (same symptoms)
>
> Let me know your thoughts.
>
> Best regards,
>
> Peter
>
> Sent from Mail<https://go.microsoft.com/fwlink/?LinkId=550986> for Windows 10
>
> From: Magnus Ihse Bursie<mailto:magnus.ihse.bursie at oracle.com>
> Sent: Wednesday, October 4, 2017 1:04 AM
> To: Peter Budai<mailto:peterbudai at hotmail.com>; build-dev at openjdk.java.net<mailto:build-dev at openjdk.java.net>
> Subject: Re: Building OpenJDK9 on MSYS2
>
> Actually, it wasn't so much remaining trouble. :-) I fired up msys2 and
> checked out where I left off. It turned out that the remaining snag was
> that msys2 tries to convert command lines automatically, from "unix"
> style paths to "windows" style paths. Unfortunately, it does not do this
> very well and it breaks all sorts of things. We already have a FIXPATH
> solution in place which deals with this, so basically all I had to do
> was disable this (by setting MSYS2_ARG_CONV_EXCL to "*"). However, this
> broke our cygpath replacement hack (!) so I had to disable it there.
> Sigh. Anyway, with those fixes it ran and worked well. (I also
> discovered and fixed a bug related to how we set up the FIXPATH variable
> on msys, but it only triggers in certain circumstances).
>
> With this patch I now jdk9 seems to build fine on msys2. It should apply
> cleanly on jdk9/jdk9. Since it turned out to be so trivial, I'll try to
> get it in in jdk10.
>
> Here's the patch if you want to apply it yourself:
>
> diff -r a08cbfc0e4ec common/autoconf/basics_windows.m4
> --- a/common/autoconf/basics_windows.m4    Thu Aug 03 18:56:56 2017 +0000
> +++ b/common/autoconf/basics_windows.m4    Wed Oct 04 00:53:58 2017 +0200
> @@ -42,7 +42,7 @@
>        windows_path=`$CYGPATH -m "$unix_path"`
>        $1="$windows_path"
>      elif test "x$OPENJDK_BUILD_OS_ENV" = "xwindows.msys"; then
> -    windows_path=`cmd //c echo $unix_path`
> +    windows_path=`MSYS2_ARG_CONV_EXCL= cmd //c echo $unix_path`
>        $1="$windows_path"
>      fi
>    ])
> @@ -136,6 +136,16 @@
>      fi
>    ])
>
> +AC_DEFUN([BASIC_MSYS_UPDATE_FIXPATH],
> +[
> +  # Take all collected prefixes and turn them into a
> -m/c/foo@/c/bar at ... command line
> +  # @ was chosen as separator to minimize risk of other tools messing
> around with it
> +  all_unique_prefixes=`echo "${all_fixpath_prefixes@<:@@@:>@}" \
> +      | tr ' ' '\n' | $GREP '^/./' | $SORT | $UNIQ`
> +  fixpath_argument_list=`echo $all_unique_prefixes  | tr ' ' '@'`
> +  FIXPATH="$FIXPATH_BIN -m$fixpath_argument_list"
> +])
> +
>    AC_DEFUN([BASIC_FIXUP_PATH_MSYS],
>    [
>      path="[$]$1"
> @@ -143,7 +153,7 @@
>      new_path="$path"
>      if test "x$has_colon" = x; then
>        # Not in mixed or Windows style, start by that.
> -    new_path=`cmd //c echo $path`
> +    new_path=`MSYS2_ARG_CONV_EXCL= cmd //c echo $path`
>      fi
>
>      BASIC_MAKE_WINDOWS_SPACE_SAFE_MSYS([$new_path])
> @@ -155,6 +165,8 @@
>
>      # Save the first 10 bytes of this path to the storage, so fixpath
> can work.
>      all_fixpath_prefixes=("${all_fixpath_prefixes@<:@@@:>@}"
> "${new_path:0:10}")
> +  # We might need to re-evaluate FIXPATH.
> +  BASIC_MSYS_UPDATE_FIXPATH
>    ])
>
>    AC_DEFUN([BASIC_FIXUP_EXECUTABLE_CYGWIN],
> @@ -293,7 +305,7 @@
>        # Do not save /bin paths to all_fixpath_prefixes!
>      else
>        # Not in mixed or Windows style, start by that.
> -    new_path=`cmd //c echo $new_path`
> +    new_path=`MSYS2_ARG_CONV_EXCL= cmd //c echo $new_path`
>        BASIC_MAKE_WINDOWS_SPACE_SAFE_MSYS([$new_path])
>        # Output is in $new_path
>        BASIC_WINDOWS_REWRITE_AS_UNIX_PATH(new_path)
> @@ -302,6 +314,8 @@
>
>        # Save the first 10 bytes of this path to the storage, so fixpath
> can work.
>        all_fixpath_prefixes=("${all_fixpath_prefixes@<:@@@:>@}"
> "${new_path:0:10}")
> +    # We might need to re-evaluate FIXPATH.
> +    BASIC_MSYS_UPDATE_FIXPATH
>      fi
>    ])
>
> @@ -347,6 +361,10 @@
>        WINDOWS_ENV_VENDOR='msys'
>        WINDOWS_ENV_VERSION="$MSYS_VERSION"
>
> +    # Prohibit msys2 path conversion from trying to be "intelligent",
> and rely
> +    # on fixpath instead.
> +    export MSYS2_ARG_CONV_EXCL="*"
> +
>        AC_MSG_CHECKING([msys root directory as unix-style path])
>        # The cmd output ends with Windows line endings (CR/LF), the grep
> command will strip that away
>        MSYS_ROOT_PATH=`cd / ; cmd /c cd | $GREP ".*"`
> @@ -391,10 +409,7 @@
>        elif test "x$OPENJDK_BUILD_OS_ENV" = xwindows.msys; then
>          # Take all collected prefixes and turn them into a
> -m/c/foo@/c/bar at ... command line
>          # @ was chosen as separator to minimize risk of other tools
> messing around with it
> -      all_unique_prefixes=`echo "${all_fixpath_prefixes@<:@@@:>@}" \
> -          | tr ' ' '\n' | $GREP '^/./' | $SORT | $UNIQ`
> -      fixpath_argument_list=`echo $all_unique_prefixes  | tr ' ' '@'`
> -      FIXPATH="$FIXPATH_BIN -m$fixpath_argument_list"
> +      BASIC_MSYS_UPDATE_FIXPATH
>        fi
>        FIXPATH_SRC_W="$FIXPATH_SRC"
>        FIXPATH_BIN_W="$FIXPATH_BIN"
> diff -r a08cbfc0e4ec common/autoconf/build-aux/config.sub
> --- a/common/autoconf/build-aux/config.sub    Thu Aug 03 18:56:56 2017 +0000
> +++ b/common/autoconf/build-aux/config.sub    Wed Oct 04 00:53:58 2017 +0200
> @@ -30,7 +30,7 @@
>    DIR=`dirname $0`
>
>    # First, filter out everything that doesn't begin with "aarch64-"
> -if ! echo $* | grep '^aarch64-' >/dev/null ; then
> +if ! echo $* | grep -e '^aarch64-' -e 'msys' >/dev/null ; then
>        . $DIR/autoconf-config.sub "$@"
>        # autoconf-config.sub exits, so we never reach here, but just in
>        # case we do:
> @@ -38,13 +38,17 @@
>    fi
>
>    while test $# -gt 0 ; do
> -    case $1 in
> +    case $1 in
>            -- )   # Stop option processing
>                shift; break ;;
>            aarch64-* )
>                config=`echo $1 | sed 's/^aarch64-/arm-/'`
>                sub_args="$sub_args $config"
>                shift; ;;
> +        *-msys )
> +            config=`echo $1 | sed 's/msys/mingw32/'`
> +            sub_args="$sub_args $config"
> +            shift; ;;
>            - )    # Use stdin as input.
>                sub_args="$sub_args $1"
>                shift; break ;;
> diff -r a08cbfc0e4ec common/autoconf/spec.gmk.in
> --- a/common/autoconf/spec.gmk.in    Thu Aug 03 18:56:56 2017 +0000
> +++ b/common/autoconf/spec.gmk.in    Wed Oct 04 00:53:58 2017 +0200
> @@ -120,6 +120,13 @@
>      # On Windows, the Visual Studio toolchain needs the PATH to be adjusted
>      # to include Visual Studio tools (this needs to be in cygwin/msys
> style).
>      export PATH:=@VS_PATH@
> +
> +endif
> +
> +ifeq ($(OPENJDK_TARGET_OS_ENV), windows.msys)
> +  # On msys2, prohibit msys path conversion from trying to be
> +  # "intelligent", and rely on fixpath instead.
> +  export MSYS2_ARG_CONV_EXCL:=*
>    endif
>
>    SYSROOT_CFLAGS := @SYSROOT_CFLAGS@
>
> /Magnus
>
> On 2017-10-03 22:34, Magnus Ihse Bursie wrote:
>> I gave msys2 a shot some time ago, but it ended up too much trouble.
>> I'll share some of my notes from that attempt, for what it's worth.
>>
>> To install package X/Y, run "pacman -S X/Y". Missing tools and
>> packages where to find them:
>> cmp: msys/diffutils
>> tar: msys/tar
>> make: msys/make
>> unzip: msys/unzip
>> zip: msys/zip
>>
>> config.sub reports msys as "x86_64-pc-mingw32" but msys2 as
>> "x86_64-pc-msys". This patch adds postprocessing in "our" config.sub
>> to report msys2 similar to msys. (Opinions, including my own :-) may
>> vary if this really is the best way..)
>>
>> diff -r b88023f46daa common/autoconf/build-aux/config.sub
>> --- a/common/autoconf/build-aux/config.sub      Fri Jan 27 10:15:41
>> 2017 +0100
>> +++ b/common/autoconf/build-aux/config.sub      Fri Feb 03 05:00:25
>> 2017 -0700
>> @@ -30,7 +30,7 @@
>>   DIR=`dirname $0`
>>
>>   # First, filter out everything that doesn't begin with "aarch64-"
>> -if ! echo $* | grep '^aarch64-' >/dev/null ; then
>> +if ! echo $* | grep -e '^aarch64-' -e 'msys' >/dev/null ; then
>>       . $DIR/autoconf-config.sub "$@"
>>       # autoconf-config.sub exits, so we never reach here, but just in
>>       # case we do:
>> @@ -45,6 +45,10 @@
>>               config=`echo $1 | sed 's/^aarch64-/arm-/'`
>>               sub_args="$sub_args $config"
>>               shift; ;;
>> +        *-msys )
>> +            config=`echo $1 | sed 's/msys/mingw32/'`
>> +            sub_args="$sub_args $config"
>> +            shift; ;;
>>           - )    # Use stdin as input.
>>               sub_args="$sub_args $1"
>>               shift; break ;;
>>
>> If I remember correctly, this got me past the configure stage at the
>> time.
>>
>> I don't think it's very hard to get it to work on msys2, I just ran
>> into one snag too many and didn't think msys2 would be used by anyone.
>>
>> /Magnus
>>
>> On 2017-10-03 17:20, Peter Budai wrote:
>>> Hello,
>>>
>>> According to
>>> http://hg.openjdk.java.net/jdk9/jdk9/file/a08cbfc0e4ec/common/doc/building.html
>>> “msys2 and the new Windows Subsystem for Linux (WSL) would likely be
>>> possible to support in a future version but that would require a
>>> community effort to implement”
>>>
>>> I’d like to help making the OpenJDK 9 build working on msys2. What is
>>> the best way to move forward? Is there a similar effort in progress?
>>>
>>> Thank you and best regards,
>>>
>>> Peter
>>>
>>>