Automake, generated source files and VPATH builds - code-generation

I'm doing VPATH builds with automake. I'm now also using generated source, with SWIG. I've got rules in Makefile.am like:
dist_noinst_DATA = whatever.swig
whatever.cpp: whatever.swig
swig -c++ -php $^
Then the file gets used later:
myprogram_SOURCES = ... whatever.cpp
It works fine when $builddir == $srcdir. But when doing VPATH builds (e.g. mkdir build; cd build; ../configure; make), I get error messages about missing whatever.cpp.
Should generated source files go to $builddir or $srcdir? (I reckon probably $builddir.)
How should dependencies and rules be specified to put generated files in the right place?

Simple answer
You should assume that $srcdir is a read-only, so you must not write anything there.
So, your generated source-code will end up in $(builddir).
By default, autotool-generated Makefiles will only look for source-files in $srcdir, so you have to tell it to check $builddir as well. Adding the following to your Makefile.am should help:
VPATH = $(srcdir) $(builddir)
After that you might end up with a no rule to make target ... error, which you should be able to fix by updating your source-generating rule as in:
$(builddir)/whatever.cpp: whatever.swig
# ...
A better solution
You might notice that in your current setup, the release tarball (as created by make dist) will contain the whatever.cpp file as part of your sources, since you added this file to the myprogram_SOURCES.
If you don't want this (e.g. because it might mean that the build-process will really take the pregenerated file rather than generating it again), you might want to use something like the following.
It uses a wrapper source-file (whatever_includer.cpp) that simply includes the generated file, and it uses -I$(builddir) to then find the generated file.
Makefile.am:
dist_noinst_DATA = whatever.swig
whatever.cpp: whatever.swig
swig -c++ -php $^
whatever_includer.cpp: whatever.cpp
myprogram_SOURCES = ... whatever_includer.cpp
myprogram_CPPFLAGS = ... -I$(builddir)
clean-local::
rm -f $(builddir)/whatever.cpp
whatever_includer.cpp:
#include "whatever.cpp"

Usually, you want to keep $srcdir readonly, so that if for instance the source is distributed unpacked on a CDROM, you can still run /.../configure from some other part of the file-system.
However if you are using SWIG to generate source code for a wrapper library, you probably want to distribute that SWIG-generated code as well so that your users do not need to install SWIG to compile your code. Then you have indeed a choice: you can decide that the SWIG-generated code should end in $builddir (it's OK: make dist will collect it there and include it in the tarball), or you could decide to output SWIG-generated code in $srcdir since it is really a source from the point of view of the distributed package. An advantage of keeping it in $srcdir is that when make distcheck attempts to build your package from a read-only source directory, it will fail on any attempt to call SWIG to regenerate the wrapper source. If you have your wrapper source in $builddir, you might not notice you have some broken rule that cause SWIG to be run on the user's host; by generating in $srcdir you ensure that SWIG is not needed by your users.
So my preference is to output SWIG wrapper sources in $srcdir. My setup for Python wrappers looks as follows:
EXTRA_DIST = spot.i
python_PYTHON = $(srcdir)/spot.py # _PYTHON is distributed by default
pyexec_LTLIBRARIES = _spot.la
MAINTAINERCLEANFILES = $(srcdir)/spot_wrap.cxx $(srcdir)/spot.py
_spot_la_SOURCES = $(srcdir)/spot_wrap.cxx $(srcdir)/spot_wrap.h
_spot_la_LDFLAGS = -avoid-version -module
_spot_la_LIBADD = $(top_builddir)/src/libspot.la
$(srcdir)/spot_wrap.cxx: $(srcdir)/spot.i
$(SWIG) -c++ -python -I$(srcdir) -I$(top_srcdir)/src $(srcdir)/spot.i
# Handle the multi-file output of SWIG.
$(srcdir)/spot.py: $(srcdir)/spot.i
$(MAKE) $(AM_MAKEFLAGS) spot_wrap.cxx
Note that I use $(srcdir) for all targets, because of limitations of the VPATH feature on various flavors of make. My setup to deal with the multiple files output by SWIG could be improved, but as these rules are not run by users and it has never caused me any problem, I do not bother.

Related

Yocto: how to configure an out-of-tree kernel module recipe that uses "inherit module"?

I have written a simple inherit module recipe to build a third-party out-of-tree kernel module called u-dma-buf:
SUMMARY = "User space mappable DMA Buffer"
DESCRIPTION = "${SUMMARY}"
LICENSE = "BSD-2-Clause"
LIC_FILES_CHKSUM = "file://LICENSE;md5=bebf0492502927bef0741aa04d1f35f5"
inherit module
SRC_URI = "git://github.com/ikwzm/udmabuf.git"
SRCREV = "9b943d49abc9c92a464e4c71e83d1c479ebbf80e"
S = "${WORKDIR}/git"
RPROVIDES_${PN} += "kernel-module-u-dma-buf"
This works correctly and generates the module file /lib/modules/[version]/extra/u-dma-buf.ko in the image.
However, looking at the docs there is an option that can be enabled: CONFIG_U_DMA_BUF_MGR that is disabled by default. If I can somehow enable this, then I can expect to find /lib/modules/[version]/extra/u-dma-buf-mgr.ko in the image also.
The project has a Kconfig file. Does bitbake have support for integrating Kbuild configuration outside of the kernel tree? If so, how do I hook into this and enable CONFIG_U_DMA_BUF_MGR? And if not, what's my best option, other than patching the Kconfig file to change the default to "y"? (EDIT: this probably won't work, as kernel sources need to be modified to incorporate Kbuild anyway - so probably a dead end unless it's a specific feature of bitbake I haven't encountered yet).
I see that the upstream Makefile also has the option CONFIG_U_DMA_BUF_MGR=m that could be used to enable this feature outside of Kbuild. I'm not sure how to pass that to the make command line though - would I need to write a custom do_compile task? Looking at the module.bbclass code, I can't see any provision for passing such an option to oe_runmake. Should I just copy/paste module_do_compile() from module.bbclass and add CONFIG_U_DMA_BUF_MGR=m? Is that the best way to do this?
So my question is, given I'm using inherit module, what is the proper way to enable this configuration option, given the recipe I have?
According to the Yocto kernel development docs:
If your module Makefile uses a different variable, you might want to override the do_compile step
That suggests to me that the correct and intended course of action in this case is to copy/paste module_do_compile() as a do_compile() override, and modify accordingly (add CONFIG_U_DMA_BUF_MGR=m):
do_compile() {
unset CFLAGS CPPFLAGS CXXFLAGS LDFLAGS
oe_runmake KERNEL_PATH=${STAGING_KERNEL_DIR} \
KERNEL_VERSION=${KERNEL_VERSION} \
CC="${KERNEL_CC}" LD="${KERNEL_LD}" \
AR="${KERNEL_AR}" \
O=${STAGING_KERNEL_BUILDDIR} \
KBUILD_EXTRA_SYMBOLS="${KBUILD_EXTRA_SYMBOLS}" \
CONFIG_U_DMA_BUF_MGR=m \
${MAKE_TARGETS}
}
Additionally, add the new module to RPROVIDES_${PN}:
RPROVIDES_${PN} += "kernel-module-u-dma-buf kernel-module-u-dma-buf-mgr"

How can I get "HelloWorld - BitBake Style" working on a newer version of Yocto?

In the book "Embedded Linux Systems with the Yocto Project", Chapter 4 contains a sample called "HelloWorld - BitBake style". I encountered a bunch of problems trying to get the old example working against the "Sumo" release 2.5.
If you're like me, the first error you encountered following the book's instructions was that you copied across bitbake.conf and got:
ERROR: ParseError at /tmp/bbhello/conf/bitbake.conf:749: Could not include required file conf/abi_version.conf
And after copying over abi_version.conf as well, you kept finding more and more cross-connected files that needed to be moved, and then some relative-path errors after that... Is there a better way?
Here's a series of steps which can allow you to bitbake nano based on the book's instructions.
Unless otherwise specified, these samples and instructions are all based on the online copy of the book's code-samples. While convenient for copy-pasting, the online resource is not totally consistent with the printed copy, and contains at least one extra bug.
Initial workspace setup
This guide assumes that you're working with Yocto release 2.5 ("sumo"), installed into /tmp/poky, and that the build environment will go into /tmp/bbhello. If you don't the Poky tools+libraries already, the easiest way is to clone it with:
$ git clone -b sumo git://git.yoctoproject.org/poky.git /tmp/poky
Then you can initialize the workspace with:
$ source /tmp/poky/oe-init-build-env /tmp/bbhello/
If you start a new terminal window, you'll need to repeat the previous command which will get get your shell environment set up again, but it should not replace any of the files created inside the workspace from the first time.
Wiring up the defaults
The oe-init-build-env script should have just created these files for you:
bbhello/conf/local.conf
bbhello/conf/templateconf.cfg
bbhello/conf/bblayers.conf
Keep these, they supersede some of the book-instructions, meaning that you should not create or have the files:
bbhello/classes/base.bbclass
bbhello/conf/bitbake.conf
Similarly, do not overwrite bbhello/conf/bblayers.conf with the book's sample. Instead, edit it to add a single line pointing to your own meta-hello folder, ex:
BBLAYERS ?= " \
${TOPDIR}/meta-hello \
/tmp/poky/meta \
/tmp/poky/meta-poky \
/tmp/poky/meta-yocto-bsp \
"
Creating the layer and recipe
Go ahead and create the following files from the book-samples:
meta-hello/conf/layer.conf
meta-hello/recipes-editor/nano/nano.bb
We'll edit these files gradually as we hit errors.
Can't find recipe error
The error:
ERROR: BBFILE_PATTERN_hello not defined
It is caused by the book-website's bbhello/meta-hello/conf/layer.conf being internally inconsistent. It uses the collection-name "hello" but on the next two lines uses _test suffixes. Just change them to _hello to match:
# Set layer search pattern and priority
BBFILE_COLLECTIONS += "hello"
BBFILE_PATTERN_hello := "^${LAYERDIR}/"
BBFILE_PRIORITY_hello = "5"
Interestingly, this error is not present in the printed copy of the book.
No license error
The error:
ERROR: /tmp/bbhello/meta-hello/recipes-editor/nano/nano.bb: This recipe does not have the LICENSE field set (nano)
ERROR: Failed to parse recipe: /tmp/bbhello/meta-hello/recipes-editor/nano/nano.bb
Can be fixed by adding a license setting with one of the values that bitbake recognizes. In this case, add a line onto nano.bb of:
LICENSE="GPLv3"
Recipe parse error
ERROR: ExpansionError during parsing /tmp/bbhello/meta-hello/recipes-editor/nano/nano.bb
[...]
bb.data_smart.ExpansionError: Failure expanding variable PV_MAJOR, expression was ${#bb.data.getVar('PV',d,1).split('.')[0]} which triggered exception AttributeError: module 'bb.data' has no attribute 'getVar'
This is fixed by updating the special python commands being used in the recipe, because #bb.data was deprecated and is now removed. Instead, replace it with #d, ex:
PV_MAJOR = "${#d.getVar('PV',d,1).split('.')[0]}"
PV_MINOR = "${#d.getVar('PV',d,1).split('.')[1]}"
License checksum failure
ERROR: nano-2.2.6-r0 do_populate_lic: QA Issue: nano: Recipe file fetches files and does not have license file information (LIC_FILES_CHKSUM) [license-checksum]
This can be fixed by adding a directive to the recipe telling it what license-info-containing file to grab, and what checksum we expect it to have.
We can follow the way the recipe generates the SRC_URI, and modify it slightly to point at the COPYING file in the same web-directory. Add this line to nano.bb:
LIC_FILES_CHKSUM = "${SITE}/v${PV_MAJOR}.${PV_MINOR}/COPYING;md5=f27defe1e96c2e1ecd4e0c9be8967949"
The MD5 checksum in this case came from manually downloading and inspecting the matching file.
Done!
Now bitbake nano ought to work, and when it is complete you should see it built nano:
/tmp/bbhello $ find ./tmp/deploy/ -name "*nano*.rpm*"
./tmp/deploy/rpm/i586/nano-dbg-2.2.6-r0.i586.rpm
./tmp/deploy/rpm/i586/nano-dev-2.2.6-r0.i586.rpm
I have recently worked on that hands-on hello world project. As far as I am concerned, I think that the source code in the book contains some bugs. Below there is a list of suggested fixes:
Inheriting native class
In fact, when you build with bitbake that you got from poky, it builds only for the target, unless you mention in your recipe that you are building for the host machine (native). You can do the latter by adding this line at the end of your recipe:
inherit native
Adding license information
It is worth mentioning that the variable LICENSE is important to be set in any recipe, otherwise bitbake rises an error. In our case, we try to build the version 2.2.6 of the nano editor, its current license is GPLv3, hence it should be mentioned as follow:
LICENSE = "GPLv3"
Using os.system calls
As the book states, you cannot dereference metadata directly from a python function. Which means it is mandatory to access metadata through the d dictionary. Bellow, there is a suggestion for the do_unpack python function, you can use its concept to code the next tasks (do_configure, do_compile):
python do_unpack() {
workdir = d.getVar("WORKDIR", True)
dl_dir = d.getVar("DL_DIR", True)
p = d.getVar("P", True)
tarball_name = os.path.join(dl_dir, p+".tar.gz")
bb.plain("Unpacking tarball")
os.system("tar -x -C " + workdir + " -f " + tarball_name)
bb.plain("tarball unpacked successfully")
}
Launching the nano editor
After successfully building your nano editor package, you can find your nano executable in the following directory in case you are using Ubuntu (arch x86_64):
./tmp/work/x86_64-linux/nano/2.2.6-r0/src/nano
Should you have any comments or questions, Don't hesitate !

Stripping base path off the unpacked source tree with bitbake SRC_URI file:// fetcher

The manual here says that there is a basepath option to SRC_URI that should "strip the specified directories from the source path when unpacking".
I'm trying to fetch the sources from a local directory, say /src/someproject.
For that purpose I configured my recipe as follows:
SRC_URI="file:///src/someproject;subdir=source;basepath=/src/someproject"
The intention was to have the sources taken from /src/someproject directory and put into build/tmp/work/target/someproject/1.0-r1/source/. Instead, I'm getting the sources under build/tmp/work/target/someproject/1.0-r1/source/src/someproject.
Is there a way to get rid of /src/someproject subdirectory inside source ?
The documentation you point to is for yocto 1.6, release April 2014. The basepath parameter appears to have been removed in later releases, with no fanfare and no apparent replacement.
Instead you can do something like:
SRCDIR = "/path/to/your/files"
SRC_URI = "file://${SRCDIR}/contents/;subdir=src"
S = "${WORKDIR}/src"
Then you can access your files under ${S}/${SRCDIR}.
If you find using the ${SRCDIR} part too cumbersome, you can hook onto the do_patch target, which need the prefunc mechanism and not _prepend if you want to use shell scripting, as do_patch is otherwise written in python:
relocate_source() {
mv ${S}/${SRCDIR}/* ${S}/
rmdir ${S}/${SRCDIR}
}
do_patch[prefunc] += "relocate_source"
This will reorganise your source before applying any patch you can add to SRC_URI.
Also note that a file:// URI does not get cached in ${DL_DIR}, so there is no name-conflict to handle (the way we would need to use downloadfilename= in an http:// URI).

Copy all files with given extension to output directory using CMake

I've seen that I can use this command in order to copy a directory using cmake:
file(COPY "myDir" DESTINATION "myDestination")
(from this post)
My problem is that I don't want to copy all of myDir, but only the .h files that are in there. I've tried with
file(COPY "myDir/*.h" DESTINATION "myDestination")
but I obtain the following error:
CMake Error at CMakeLists.txt:23 (file):
file COPY cannot find
"/full/path/to/myDIR/*.h".
How can I filter the files that I want to copy to a destination folder?
I've found the solution by myself:
file(GLOB MY_PUBLIC_HEADERS
"myDir/*.h"
)
file(COPY ${MY_PUBLIC_HEADERS} DESTINATION myDestination)
this also works for me:
install(DIRECTORY "myDir/"
DESTINATION "myDestination"
FILES_MATCHING PATTERN "*.h" )
The alternative approach provided by jepessen does not take into account the fact that sometimes the number of files to be copied is too high. I encountered the issue when doing such thing (more than 110 files)
Due to a limitation on Windows on the number of characters (2047 or 8191) in a single command line, this approach may randomly fail depending on the number of headers that are in the folder. More info here https://support.microsoft.com/en-gb/help/830473/command-prompt-cmd-exe-command-line-string-limitation
Here is my solution:
file(GLOB MY_HEADERS myDir/*.h)
foreach(CurrentHeaderFile IN LISTS MY_HEADERS)
add_custom_command(
TARGET MyTarget PRE_BUILD
COMMAND ${CMAKE_COMMAND} -E copy_if_different ${CurrentHeaderFile} ${myDestination}
COMMENT "Copying header: ${CurrentHeaderFile}")
endforeach()
This works like a charm on MacOS. However, if you have another target that depends on MyTarget and needs to use these headers, you may have some compile errors due to not found includes on Windows. Therefore you may want to prefer the following option that defines an intermediate target.
function (CopyFile ORIGINAL_TARGET FILE_PATH COPY_OUTPUT_DIRECTORY)
# Copy to the disk at build time so that when the header file changes, it is detected by the build system.
set(input ${FILE_PATH})
get_filename_component(file_name ${FILE_PATH} NAME)
set(output ${COPY_OUTPUT_DIRECTORY}/${file_name})
set(copyTarget ${ORIGINAL_TARGET}-${file_name})
add_custom_target(${copyTarget} DEPENDS ${output})
add_dependencies(${ORIGINAL_TARGET} ${copyTarget})
add_custom_command(
DEPENDS ${input}
OUTPUT ${output}
COMMAND ${CMAKE_COMMAND} -E copy_if_different ${input} ${output}
COMMENT "Copying file to ${output}."
)
endfunction ()
foreach(HeaderFile IN LISTS MY_HEADERS)
CopyFile(MyTarget ${HeaderFile} ${myDestination})
endforeach()
The downside indeed is that you end up with multiple target (one per copied file) but they should all end up together (alphabetically) since they start with the same prefix ORIGINAL_TARGET -> "MyTarget"

Is there a way to tell django compressor to create source maps

I want to be able to debug minified compressed javascript code on my production site. Our site uses django compressor to create minified and compressed js files. I read recently about chrome being able to use source maps to help debug such javascript. However I don't know how/if possible to tell the django compressor to create source maps when compressing the js files
I don't have a good answer regarding outputting separate source map files, however I was able to get inline working.
Prior to adding source maps my settings.py file used the following precompilers
COMPRESS_PRECOMPILERS = (
('text/coffeescript', 'coffee --compile --stdio'),
('text/less', 'lessc {infile} {outfile}'),
('text/x-sass', 'sass {infile} {outfile}'),
('text/x-scss', 'sass --scss {infile} {outfile}'),
('text/stylus', 'stylus < {infile} > {outfile}'),
)
After a quick
$ lessc --help
You find out you can put the less and map files in to the output css file. So my new text/less precompiler entry looks like
('text/less', 'lessc --source-map-less-inline --source-map-map-inline {infile} {outfile}'),
Hope this helps.
Edit: Forgot to add, lessc >= 1.5.0 required for this, to upgrade use
$ [sudo] npm update -g less
While I couldn't get this to work with django-compressor (though it should be possible, I think I just had issues getting the app set up correctly), I was able to get it working with django-assets.
You'll need to add the appropriate command-line argument to the less filter source code as follows:
diff --git a/src/webassets/filter/less.py b/src/webassets/filter/less.py
index eb40658..a75f191 100644
--- a/src/webassets/filter/less.py
+++ b/src/webassets/filter/less.py
## -80,4 +80,4 ## class Less(ExternalTool):
def input(self, in_, out, source_path, **kw):
# Set working directory to the source file so that includes are found
with working_directory(filename=source_path):
- self.subprocess([self.less or 'lessc', '-'], out, in_)
+ self.subprocess([self.less or 'lessc', '--line-numbers=mediaquery', '-'], out, in_)
Aside from that tiny addition:
make sure you've got the node -- not the ruby gem -- less compiler (>=1.3.2 IIRC) available in your path.
turn on the sass source-maps option buried away in chrome's web inspector config pages. (yes, 'sass' not less: less tweaked their debug-info format to match sass's since since sass had already implemented a chrome-compatible mapping and their formats weren't that different to begin with anyway...)
Not out of the box but you can extend a custom filter:
from compressor.filters import CompilerFilter
class UglifyJSFilter(CompilerFilter):
command = "uglifyjs -c -m " /
"--source-map-root={relroot}/ " /
"--source-map-url={name}.map.js" /
"--source-map={relpath}/{name}.map.js -o {output}"