Patchwork [External] Re: crosstool-ng 1.19.0 fails to build i686-unknown-mingw32

login
register
mail settings
Submitter Soloff, Steven M Belcan Engineering Group, Cincinnati, Ohio
Date Jan. 30, 2014, 6:15 p.m.
Message ID <4F66624F960F3746BB0263581CC9CE83040B7F1C@PUSMNEI1.pwemail.us>
Download mbox | patch
Permalink /patch/315426/
State New
Headers show

Comments

Hi Ray,

My mistake...  You're correct, the pass-1 compiler is successfully built
if sysroot/mingw/include exists.  I was just careless when examining the
log; the missing header error I was referring to occurs during the build
of the pass-2 compiler:

[ALL  ]    In file included from
/home/xbe1648/Documents/crosstool/i686-unknown-mingw32/.build/src/gcc-4.
8.1/libgcc/libgcc2.c:27:0:
[ERROR]
/home/xbe1648/Documents/crosstool/i686-unknown-mingw32/.build/src/gcc-4.
8.1/libgcc/../gcc/tsystem.h:87:19: fatal error: stdio.h: No such file or
directory
[ALL  ]     #include <stdio.h>
[ALL  ]                       ^
[ALL  ]    compilation terminated.

However, that error's a red herring.  My problem was that I didn't
correctly create sysroot/mingw/include.  I didn't notice the comment in
scripts/build/libc/mingw.sh regarding the creation of the symbolic link
between sysroot/mingw and sysroot/usr/${CT_TARGET}.

So instead of simply executing the following command after the build
started (which lead to the compilation error above):

	mkdir -p
/home/user/x-tools/i686-unknown-mingw32/i686-unknown-mingw32/sysroot/min
gw/include

I used the following two commands:

	mkdir -p
/home/user/x-tools/i686-unknown-mingw32/i686-unknown-mingw32/sysroot/usr
/i686-unknown-mingw32/include
	ln -sv usr/i686-unknown-mingw32
/home/user/x-tools/i686-unknown-mingw32/i686-unknown-mingw32/sysroot/min
gw

That apparently worked such that the link was pointing to the correct
location when it came time to build the pass-2 compiler.

With that said, I reformulated the patch to only perform the above two
actions at the end of do_libc_extract() instead of installing the C
library:

 do_libc_check_config() {
@@ -48,11 +58,6 @@ do_libc_start_files() {
     
     CT_Popd
 
-    # It seems mingw is strangely set up to look into /mingw instead of
-    # /usr (notably when looking for the headers). This symlink is
-    # here to workaround this, and seems to be here to last... :-/
-    CT_DoExecLog ALL ln -sv "usr/${CT_TARGET}"
"${CT_SYSROOT_DIR}/mingw"
-
     CT_EndStep
 }

It still feels like I'm breaking cohesion by doing this in
do_libc_extract().  Any suggestions for improvement would be
appreciated.

Thanks,
Steve


-----Original Message-----
From: Ray Donnelly [mailto:mingw.android@gmail.com] 
Sent: Thursday, January 30, 2014 9:44 AM
To: Soloff, Steven M Belcan Engineering Group, Cincinnati, Ohio
Cc: crossgcc@sourceware.org
Subject: [External] Re: crosstool-ng 1.19.0 fails to build
i686-unknown-mingw32

I've never needed the crt headers installed during pass-1 GCC, just that
the folder exists.

On Thu, Jan 30, 2014 at 2:07 PM, Soloff, Steven M          Belcan
Engineering Group, Cincinnati, Ohio <Steven.Soloff@pw.utc.com> wrote:
> This apparently broke beginning with GCC 4.6.0.  I didn't track down 
> the root cause of the change in the GCC build process that required 
> the sysroot files to be accessible during the pass-1 compiler build.
> However, the existence of the sysroot/mingw directory is not 
> sufficient
> -- the headers need to be there, as well.
>
> Taking a cue from the patch referenced by Drew, I instead modified 
> scripts/build/libc/mingw.sh by moving the contents of
> do_libc_start_files() to the end of do_libc_extract() [1], which 
> should only affect MinGW targets.  I verified I could build a MinGW 
> target using GCC 4.5.3 (i686) and 4.8.1 (i686 and x86_64) using a 
> patched crosstool-ng 1.19.0.
>
> However, I doubt this is the optimal solution.  Seeing the "Installing

> C library headers" step during the "Extracting and patching toolchain 
> components" step just smells bad.  I'd appreciate any suggestions from

> the devs for a better fix that I could submit.
>
> Thanks,
> Steve
>
> [1]
> --- a/scripts/build/libc/mingw.sh
> +++ b/scripts/build/libc/mingw.sh
> @@ -11,13 +11,7 @@ do_libc_extract() {
>      CT_Pushd "${CT_SRC_DIR}/mingw-w64-v${CT_WINAPI_VERSION}/"
>      CT_Patch nochdir mingw-w64 "${CT_WINAPI_VERSION}"
>      CT_Popd
> -}
>
> -do_libc_check_config() {
> -    :
> -}
> -
> -do_libc_start_files() {
>      local -a sdk_opts
>
>      CT_DoStep INFO "Installing C library headers"
> @@ -56,6 +50,14 @@ do_libc_start_files() {
>      CT_EndStep
>  }
>
> +do_libc_check_config() {
> +    :
> +}
> +
> +do_libc_start_files() {
> +    :
> +}
> +
>  do_libc() {
>      CT_DoStep INFO "Building mingw-w64 files"
>
>
> -----Original Message-----
> From: Drew Bliss [mailto:drewb@valvesoftware.com]
> Sent: Monday, January 20, 2014 7:32 PM
> To: 'crossgcc@sourceware.org'
> Subject: RE: crosstool-ng 1.19.0 fails to build i686-unknown-mingw32
>
> The x-tools directory appears to be cleaned up and recreated when 
> running ct-ng build, so any manually created directory is gone by the 
> time it gets to the point of needing it.
>
> Back in Aug 2012 the same issue was reported with 1.16 and a user said

> that this patch reordering steps fixed the problem.  It doesn't appear

> to fix it with 1.19, plus the patch was considered undesirable since 
> it breaks non-mingw builds.
>
> --- a/lib/ct-ng.1.16.0/steps.mk 2012-08-06 14:55:00.848409671 +0200
> +++ b/lib/ct-ng.1.16.0/steps.mk 2012-08-06 14:39:24.736900247 +0200
> @@ -36,9 +36,9 @@
>              binutils_for_host   \
>              elf2flt_for_host    \
>              sstrip_for_host     \
> +            libc_start_files    \
>              cc_core_pass_1      \
>              kernel_headers      \
> -            libc_start_files    \
>              cc_core_pass_2      \
>              libc                \
>              cc_for_build        \
>
> -----Original Message-----
> From: Ray Donnelly [mailto:mingw.android@gmail.com]
> Sent: Monday, January 20, 2014 4:19 PM
> To: Drew Bliss
> Cc: crossgcc@sourceware.org
> Subject: Re: crosstool-ng 1.19.0 fails to build i686-unknown-mingw32
>
> mkdir -p .. of course!
>
> On Tue, Jan 21, 2014 at 12:18 AM, Ray Donnelly 
> <mingw.android@gmail.com>
> wrote:
>> As a workaround you can make the parent directory before commencing
> the build:
>>
>> mkdir
>> /home/user/x-tools/i686-unknown-mingw32/i686-unknown-mingw32/sysroot/
>> m
>> ingw
>>
>>
>> On Mon, Jan 20, 2014 at 11:50 PM, Drew Bliss 
>> <drewb@valvesoftware.com>
> wrote:
>>> This is the same issue that Daniel Pauli mailed about in Mar 2013 
>>> but
> there didn't seem to be a reply.  Does anybody know what the problem 
> is here?
>>>
>>> From: Daniel Pauli <dapaulid at gmail dot com>
>>> To: crossgcc at sourceware dot org
>>> Date: Wed, 6 Mar 2013 19:47:25 +0100
>>> Subject: crosstool-ng 1.18.0 fails to build i686-unknown-mingw32 for

>>> gcc 4.7.2 Hey there
>>>
>>> I'm trying to build the i686-unknown-mingw32 sample for gcc 4.7.2 
>>> but
>
>>> it fails at the following step:
>>>
>>> [ALL  ]    The directory that should contain system headers does not
> exist:
>>> [ALL  ]
> /home/user/x-tools/i686-unknown-mingw32/i686-unknown-mingw32/sysroot/m
> in
> gw/include
>>> [ERROR]    make[2]: *** [stmp-fixinc] Error 1
>>> [ALL  ]    rm gcc.pod
>>> [ALL  ]    make[2]: Leaving directory
>>>
> `/home/user/crosstool-ng-1.18.0/test/.build/i686-unknown-mingw32/build
> /b
> uild-cc-core-pass-1/gcc'
>>> [ERROR]    make[1]: *** [all-gcc] Error 2
>>> [ALL  ]    make[1]: Leaving directory
>>>
> `/home/user/crosstool-ng-1.18.0/test/.build/i686-unknown-mingw32/build
> /b
> uild-cc-core-pass-1'
>>> [ERROR]
>>> [ERROR]  >>
>>> [ERROR]  >>  Build failed in step 'Installing pass-1 core C
compiler'
>>> [ERROR]  >>        called in step '(top-level)'
>>> [ERROR]  >>
>>> [ERROR]  >>  Error happened in: CT_DoExecLog[scripts/functions@257]
>>> [ERROR]  >>        called from:
> do_cc_core_backend[scripts/build/cc/gcc.sh@448]
>>> [ERROR]  >>        called from:
> do_cc_core_pass_1[scripts/build/cc/gcc.sh@101]
>>> [ERROR]  >>        called from: main[scripts/crosstool-NG.sh@632]
>>>
>>>
>>> --
>>> For unsubscribe information see http://sourceware.org/lists.html#faq
>>>
>
> --
> For unsubscribe information see http://sourceware.org/lists.html#faq
>

--
For unsubscribe information see http://sourceware.org/lists.html#faq

Patch

--- a/scripts/build/libc/mingw.sh
+++ b/scripts/build/libc/mingw.sh
@@ -11,6 +11,16 @@  do_libc_extract() {
     CT_Pushd "${CT_SRC_DIR}/mingw-w64-v${CT_WINAPI_VERSION}/"
     CT_Patch nochdir mingw-w64 "${CT_WINAPI_VERSION}"
     CT_Popd
+
+    # Beginning with GCC 4.6.0, the "sysroot/mingw/include" directory
+    # must exist before the pass-1 compiler is built.
+    CT_mkdir_pushd "${CT_SYSROOT_DIR}/usr/${CT_TARGET}/include"
+    CT_Popd
+
+    # It seems mingw is strangely set up to look into /mingw instead of
+    # /usr (notably when looking for the headers). This symlink is
+    # here to workaround this, and seems to be here to last... :-/
+    CT_DoExecLog ALL ln -sv "usr/${CT_TARGET}"
"${CT_SYSROOT_DIR}/mingw"
 }