pacakge/qt5/qt5base: fix build with ccache

Building with ccache failed with:

Running configuration tests...
Failed to process makespec for platform 'devices/linux-buildroot-g++'
Project ERROR: Compiler <path_to_output_dir>/host/usr/bin/ccache <path_to_output_dir>/host/usr/bin/<cross_compile>-g++ not found. Check the value of CROSS_COMPILE -device-option Could not read qmake configuration file <path_to_output_dir>/build/qt5base-5.5.0/mkspecs/devices/linux-buildroot-g++/qmake.conf.
Error processing project file: /dev/null

This was caused by Buildroot setting this in qt5base-5.5.0/mkspecs/devices/linux-buildroot-g++/qmake.conf:

    QMAKE_CXX               = $${BR_CCACHE} $${CROSS_COMPILE}g++

But qt5base-5.5.0/mkspecs/features/device_config.prf expects QMAKE_CXX to be a valid (absolute or QMAKE_PATH_ENV-relative) path to an existing file, which is not possible if using ccache as above.

Add a patch removing these tests since the toolchain is already tested by Buildroot, which allows to keep using ccache.

Signed-off-by Beno?t Th?baudeau <[email protected]>

---
 .../qt5/qt5base/0009-fix-build-with-ccache.patch   | 60 ++++++++++++++++++++++
 1 file changed, 60 insertions(+)
 create mode 100644 package/qt5/qt5base/0009-fix-build-with-ccache.patch

package/qt5/qt5base/0009-fix-build-with-ccache.patch new file +45 -0

new file mode 100644
index 0000000..04d41a4
@@ -0,0 +1,60 @@
+Fix build with ccache
+
+Building with Buildroot and ccache failed with:
+
+    Running configuration tests...
+    Failed to process makespec for platform 'devices/linux-buildroot-g++'
+    Project ERROR: Compiler <path_to_output_dir>/host/usr/bin/ccache <path_to_output_dir>/host/usr/bin/<cross_compile>-g++ not found. Check the value of CROSS_COMPILE -device-option
+    Could not read qmake configuration file <path_to_output_dir>/build/qt5base-5.5.0/mkspecs/devices/linux-buildroot-g++/qmake.conf.
+    Error processing project file: /dev/null
+
+This was caused by Buildroot setting this in
+qt5base-5.5.0/mkspecs/devices/linux-buildroot-g++/qmake.conf:
+
+    QMAKE_CXX               = $${BR_CCACHE} $${CROSS_COMPILE}g++
+
+But qt5base-5.5.0/mkspecs/features/device_config.prf expects QMAKE_CXX
+to be a valid (absolute or QMAKE_PATH_ENV-relative) path to an existing
+file, which is not possible if using ccache as above.
+
+Remove these tests since the toolchain is already tested by Buildroot,
+which allows to keep using ccache.
+
+Signed-off-by: Beno?t Th?baudeau <[email protected]>
+---
+ mkspecs/features/device_config.prf | 23 -----------------------
+ 1 file changed, 23 deletions(-)
+
+diff --git a/mkspecs/features/device_config.prf b/mkspecs/features/device_config.prf
+index cd3a0cf..ef16540 100644
+--- a/mkspecs/features/device_config.prf
++++ b/mkspecs/features/device_config.prf
+@@ -14,28 +14,5 @@ host_build {
+ 
+ # Provide a function to be used by mkspecs
+ defineTest(deviceSanityCheckCompiler) {
+-    equals(QMAKE_HOST.os, Windows): \
+-        sfx = .exe
+-    else: \
+-        sfx =
+-
+-    # Check if the binary exists with an absolute path. Do this check
+-    # before the CROSS_COMPILE empty check below to allow the mkspec
+-    # to derive the compiler path from other device options.
+-    exists($$QMAKE_CXX$$sfx):return()
+-

Quoted 6 lines · Beno?t Th?baudeau

+-    # Check if QMAKE_CXX points to an executable.
+-    ensurePathEnv()
+-    for (dir, QMAKE_PATH_ENV) {
+-        exists($$dir/$${QMAKE_CXX}$$sfx): \
+-            return()
+-    }

Instead of removing this entire piece of code, can you try to replace just this last part by something like:

        system($${QMAKE_CXX}$$sfx --version): return()

Instead of trying to look for a file, it will actually try to run the compiler.

Also, can you report the bug and submit the patch upstream to Qt? See
https://bugreports.qt.io/secure/Dashboard.jspa.

Quoted 11 lines · Thomas Petazzoni

Quoted 6 lines · Beno?t Th?baudeau

+-    # Check if QMAKE_CXX points to an executable.
+-    ensurePathEnv()
+-    for (dir, QMAKE_PATH_ENV) {
+-        exists($$dir/$${QMAKE_CXX}$$sfx): \
+-            return()
+-    }

Instead of removing this entire piece of code, can you try to replace just this last part by something like:

Quoted 7 lines

        system($${QMAKE_CXX}$$sfx --version): return()

Instead of trying to look for a file, it will actually try to run the compiler.

Also, can you report the bug and submit the patch upstream to Qt? See
https://bugreports.qt.io/secure/Dashboard.jspa.

I have opened a bug report upstream as you suggested:
https://bugreports.qt.io/browse/QTBUG-47951

As explained in this report, I think that your suggestion would work for Buildroot (not yet tested), but not for upstream Qt.

My point with this patch was that upstream could consider this Buildroot use case as illegal because the definition of QMAKE_CXX
(http://doc.qt.io/qt-5/qmake-variable-reference.html#qmake-cxx) only mentions a filename and a path as possibilities. In that case, the bug is in Buildroot, and since this sanity check of the compiler is not needed for Buildroot, we can just remove this test. If we don't want a local patch for this, maybe we could use a wrapper above ccache, or put ccache into QMAKE_CXX and the compiler into QMAKE_CXXFLAGS (not tested, and a bit ugly).

Wait and see what upstream says.

Quoted 13 lines

As explained in this report, I think that your suggestion would work for Buildroot (not yet tested), but not for upstream Qt.

My point with this patch was that upstream could consider this Buildroot use case as illegal because the definition of QMAKE_CXX
(http://doc.qt.io/qt-5/qmake-variable-reference.html#qmake-cxx) only mentions a filename and a path as possibilities. In that case, the bug is in Buildroot, and since this sanity check of the compiler is not needed for Buildroot, we can just remove this test. If we don't want a local patch for this, maybe we could use a wrapper above ccache, or put ccache into QMAKE_CXX and the compiler into QMAKE_CXXFLAGS (not tested, and a bit ugly).

Quoted 1 line

Wait and see what upstream says.

Upstream recommends this approach:
http://lists.qt-project.org/pipermail/qt-creator/2014-January/003148.html

What do you think?

Quoted 13 lines · Beno?t Th?baudeau

Quoted 13 lines

As explained in this report, I think that your suggestion would work for Buildroot (not yet tested), but not for upstream Qt.

My point with this patch was that upstream could consider this Buildroot use case as illegal because the definition of QMAKE_CXX
(http://doc.qt.io/qt-5/qmake-variable-reference.html#qmake-cxx) only mentions a filename and a path as possibilities. In that case, the bug is in Buildroot, and since this sanity check of the compiler is not needed for Buildroot, we can just remove this test. If we don't want a local patch for this, maybe we could use a wrapper above ccache, or put ccache into QMAKE_CXX and the compiler into QMAKE_CXXFLAGS (not tested, and a bit ugly).

Quoted 6 lines

Quoted 1 line

Wait and see what upstream says.

Upstream recommends this approach:
http://lists.qt-project.org/pipermail/qt-creator/2014-January/003148.html

What do you think?

Well, we already have a wrapper for the compiler in the case of the external toolchain. And we will probably also extend the wrapper in the future to the internal toolchain case, in order to support per-package staging directory. In this case, we could take advantage of this wrapper to include the ccache functionality, and avoid having CC and CXX values that have two words.

I really would like to have the wrapper suggested in this e-mail calling our wrapper, then calling the real compiler.

OK, but in the meantime, until we have a wrapper for the internal toolchain too, what do you want to do to fix the current build failure?
- Forbid the use of ccache with Qt 5?
- Use the current patch?
- Use the wrapper from the URL above? In that case, where would you put it in Buildroot (qt5base, new package, toolchain, support)?

Quoted 6 lines · Beno?t Th?baudeau

OK, but in the meantime, until we have a wrapper for the internal toolchain too, what do you want to do to fix the current build failure?
- Forbid the use of ccache with Qt 5?
- Use the current patch?
- Use the wrapper from the URL above? In that case, where would you put it in Buildroot (qt5base, new package, toolchain, support)?

For 2015.08, your patch is certainly the simplest solution.

OK, but in the meantime, until we have a wrapper for the internal toolchain too, what do you want to do to fix the current build failure?
- Forbid the use of ccache with Qt 5?

- Use the current patch?

Quoted 3 lines · Beno?t Th?baudeau

OK, but in the meantime, until we have a wrapper for the internal toolchain too, what do you want to do to fix the current build failure?
- Forbid the use of ccache with Qt 5?

I would not recommend that. Qt5base is *huge* and recompiling it over and over again (as I do when I am developing with/against buildroot) is just too time consuming.

Quoted 1 line

- Use the current patch?

You can also look at one of my previous patches (see list archive). I simply don't have the time to respond to all correction requests unfortunately.