autoconf: how do I substitute the library prefix? - makefile

CLISP's interface to PARI is configured with the configure.in containing AC_LIB_LINKFLAGS([pari]) from lib-link.m4.
The build process also requires the Makefile to know where the datadir of PARI is located. To this end, Makefile.in has
prefix = #LIBPARI_PREFIX#
DATADIR = #datadir#
and expects to find $(DATADIR)/pari/pari.desc (normally
/usr/share/pari/pari.desc or /usr/local/share/pari/pari.desc).
This seems to work on Mac OS X where PARI is installed by homebrew in /usr/local (and LIBPARI_PREFIX=/usr/local), but not on Ubuntu, where PARI is in /usr, and LIBPARI_PREFIX is empty.
How do I insert the location of the PARI's datadir into the Makefile?
PS. I also asked this on the autoconf mailing list.
PPS. In response to #BrunoHaible's suggestion, here is the meager attempt at debugging on Linux (where LIBPARI_PREFIX is empty).
$ bash -x configure 2>&1 | grep found_dir
+ found_dir=
+ eval ac_val=$found_dir
+ eval ac_val=$found_dir

You are trying to use $(prefix) in an unintended way. In an Autotools-based build system, the $(prefix) represents a prefix to the target installation location of the software you're building. By setting it in your Makefile.in, you are overriding the prefix that configure will try to assign. However, since you appear not to have any installation targets anyway, at least at that level, that's probably more an issue of poor form than a cause for malfunction.
How do I insert the location of the PARI's datadir into the Makefile?
I'd recommend computing or discovering the needed directory in your configure script, and exporting it to the generated Makefile via its own output variable. Let's take the second part first, since it's simple. In configure.in, having in some manner located the wanted data directory and assigned it to a variable
DATADIR=...
, you would make an output variable of that via the AC_SUBST macro:
AC_SUBST([DATADIR])
Since you are using only Autoconf, not Automake, you would then manually receive that into your Makefile by changing the assignment in your Makefile.in:
DATDIR = #DATADIR#
Now, as for locating the data directory in the first place, you have to know what you're trying to implement before you can implement it. From your question and followup comments, it seems to me that you want this:
Use a data directory explicitly specified by the user if there is one. Otherwise,
look for a data directory relative to the location of the shared library. If it's not found there then
(optional) look under the prefix specified to configure, or specifically in the specified datadir (both of which may come from the top-level configure). Finally, if it still has not been found then
look in some standard locations.
To create a configure option by which the user can specify a custom data directory, you would probably use the AC_ARG_WITH macro, maybe like this:
AC_ARG_WITH([pari-datadir], [AS_HELP_STRING([--with-pari-datadir],
[explicitly specifies the PARI data directory])],
[], [with_pari_datadir=''])
Thanks to #BrunoHaible, we see that although the Gnulib manual does not document it, the macro's internal documentation specifies that if AC_LIB_LINKFLAGS locates libpari then it will set LIBPARI_PREFIX to the library directory prefix. You find that that does work when the --with-libpari option is used to give it an alternative location to search, so I suggest working with that. You certainly can try to debug AC_LIB_LINKFLAGS to make it set LIBPARI_PREFIX in all cases in which the lib is found, but if you don't want to go to that effort then you can work around it (see below).
Although the default or specified installation prefix is accessible in configure as $prefix, I would suggest instead going to the specified $datadir. That is slightly tricky, however, because by default it refers to the prefix indirectly. Thus, you might do this:
eval "datadir_expanded=${datadir}"
Finally, you might hardcode a set of prefixes such as /usr and /usr/local.
Following on from all the foregoing, then, your configure.in might do something like this:
DATADIR=
for d in \
${with_pari_datadir} \
${LIBPARI_PREFIX:+${LIBPARI_PREFIX}/share/pari} \
${datadir_expanded}/pari \
/usr/local/share/pari \
/usr/share/pari
do
AS_IF([test -r "$[]d/pari.desc"], [DATADIR="$[]d"; break])
done
AS_IF([test x = "x$DATADIR"], [AC_MSG_ERROR(["Could not identify PARI data directory"])])
AC_SUBST([DATADIR])

Instead of guessing the location of datadir, why don't you ask PARI/GP where its datadir is located? Namely,
$ echo "default(datadir)" | gp -qf
"/usr/share/pari"
does the trick.

Related

Patch bitbake to use custom `wpa_supplicant.conf`

I have a wpa_supplicant.conf that I want to use at first boot. I tried to patch poky's recipe as follows. This is my wpa_supplicant_2.6.bbappend:
FILESEXTRAPATHS_append := ":${THISDIR}/${PN}"
SRC_URI_append = " file://wpa_supplicant.conf"
I have the conf file stored relative to the bbappend in wpa_supplicant/wpa_supplicant.conf. Still the original poky conf file is always added to the image.
How can I inject my config file?
(I'm on branch warrior)
NOTE/EDIT: For quick and efficient problem solving I recommend also discussing issues in the IRC channel (as also happened in this case). Super helpful people there!
First, the recipe is called wpa-supplicant and not wpa_supplicant so you need to name the bbappend wpa-supplicant_2.6.bbapend and not wpa_supplicant_2.6.bbappend. Remember, no underscore, no uppercase letter in recipe or package name.
One can check that a bbappend is parsed by using bitbake-layers show-appends wpa-supplicant. You'd have seen that your bbappend was not taken into account.
Then, one should usually use FILESEXTRAPATHS_prepend := because you want your path to be traversed before all the other ones.
FILESEXTRAPATHS_prepend := "${THISDIR}/${PN}:"
Finally, since the recipe is named wpa-supplicant and not wpa_supplicant, ${PN} will be wpa-supplicant, thus you need to put your file in wpa-supplicant/ and not wpa_supplicant.
To check in which order paths are traversed for files, one can run bitbake -e wpa-supplicant | grep -e "^FILESPATH=", the paths are traversed from leftmost to rightmost. The first file which matches the full path will be taken.
After discussing on IRC with you, I can add that we also figured out that wpa_supplicant.conf is installed in the documentation directory but wpa_supplicant.conf-sane is the one that will be used as wpa_supplicant.conf in the final image for the target. So one would need to name the wpa_supplicant.conf file as wpa_supplicant.conf-sane in order for it to replace the wpa_supplicant.conf file in the final image.
c.f. https://git.yoctoproject.org/cgit.cgi/poky/tree/meta/recipes-connectivity/wpa-supplicant/wpa-supplicant_2.6.bb?h=thud#n88 and https://git.yoctoproject.org/cgit.cgi/poky/tree/meta/recipes-connectivity/wpa-supplicant/wpa-supplicant_2.6.bb?h=thud#n91

OSX Yosemite - invalid characters in path

I'm having trouble installing Homebrew, and it looks like the problem (invalid byte sequence in UTF-8 (ArgumentError) error message) is due to a fault in my PATH. when I run echo $PATH, I get the following (be sure to scroll to the far right in the code block below to see the fault):
/usr/local/mysql/bin:/opt/subversion/bin/:/usr/local/bin:/usr/local/sbin:/usr/bin:/bin:/usr/sbin:/sbin:PATH=/usr/local/share/python:$PATH:PATH=/usr/local/share/npm/bin:$PATH:export PATH��:/usr/local/git/bin
I have no idea where those strange characters just before the /usr/local/git/bin path are coming from. I'm using bash (tried zshell, same problem). I can't find where this coming from. my .bash_profile file looks clean, too.
Examine your ~/.profile, ~/.bash_profile, and ~/.bashrc files. It is also possible that these values are being set in /etc/profile or other system-wide files, but the nature of the errors is such that I would assume they are the result of your own manual edits of your personal files. You're the best judge to decide what you may have changed and when.
Look for entries such as the following:
PATH='/usr/local/share/python:$PATH'
PATH='PATH=/usr/local/share/npm/bin:$PATH'
PATH='export PATH��:/usr/local/git/bin'
They should instead look like
PATH=/usr/local/share/python:$PATH
PATH=/usr/local/share/npm/bin:$PATH
PATH=$PATH:/usr/local/git/bin
# Not strictly necessary -- PATH is already exported
export PATH
See a pattern? These all add a value to PATH by setting it to prefix:$PATH or $PATH:suffix where the new prefix or suffix is the directory where you have installed a local package. Whether the new value should be a prefix or a suffix is probably of little consequence -- except if you are trying to override a system version (say, your /usr/local/git is installed because /usr/bin/git is too old -- then, it's important that your local version comes first in the PATH). There is also a minuscule speed difference the first time you run a command; the shell has to look in more places before finding it if it's at the end of the PATH.
The difference in quoting is significant -- PATH=$PATH:suffix expands to the old value of PATH with :suffix added on, whereas PATH='$PATH:suffix' replaces the old value of PATH with the literal string $PATH with :suffix added on. You're lucky none of this has ended up replacing your original PATH completely, which would render your shell by and large inoperable (at least until you discover that you have to use /bin/ls instead of just ls).

automake: how set path to linker script?

I've just set up a cross-helloworld automake project (for stm32f4-discovery). There I have a custom discovery.ld scrpt. I put a line in my Makefile.amAM_LDFLAGS = -T discovery.ld. The problem starts when I run confgure from a different folder (e.g. /path/to/source/build) to the one in which the script is situated (/path/to/source). Effectively, being in /path/to/source/build directory, make runs gcc -T discovery.ld ... and fails to find the script because it's in /path/to/source directory and it's not included in the search path list.
-L/path/to/source or -L.. would solve the problem but I don't want to hardcode things.
Maybe there a autoconf/automake variable exists which points to the folder where configure script (and also my discovery.ld) are situated so that I could use it in my Makefile.am?
I'd be glad to any advice.
Many thanks to William Pursell:
AM_LDFLAGS = -T $(top_srcdir)/path/to/discovery.ld

Organizing asset files in a Go project

I have a project that contains a folder to manage file templates, but it doesn't look like Go provides any support for non-Go-code project files. The project itself compiles to an executable, but it needs to know where this template folder is in order to operate correctly. Right now I do a search for $GOPATH/src/<templates>/templates, but this feels like kind of a hack to me because it would break if I decided to rename the package or host it somewhere else.
I've done some searching and it looks like a number of people are interested in being able to "compile" the asset files by embedding them in the final binary, but I'm not sure how I feel about this approach.
Any ideas?
Either pick a path (or a list of paths) that users are expected to put the supporting data in (/usr/local/share/myapp, ...) or just compile it into the binary.
It depends on how you are planning to distribute the program. As a package? With an installer?
Most of my programs I enjoy just having a single file to deploy and I just have a few templates to include, so I do that.
I have an example using go-bindata where I build the html template with a Makefile, but if I build with the 'devel' flag it will read the file at runtime instead to make development easier.
I can think of two options, use a cwd flag, or infer from cwd and arg 0:
-cwd path/to/assets
path/to/exe -cwd=$(path/to/exe/assets)
Internally, the exectable would chdir to wherever cwd points to, and then it can use relative paths throughout the application. This has the added benefit that the user can change the assets without having to recompile the program.
I do this for config files. Basically the order goes:
process cmd arguments, looking for a -cwd variable (it defaults to empty)
chdir to -cwd
parse config file
reparse cmd arguments, overwriting the settings in the config file
I'm not sure how many arguments your app has, but I've found this to be very useful, especially since Go doesn't have a standard packaging tool that will compile these assets in.
infer from arg 0
Another option is to use the first argument and get the path to the executable. Something like this:
here := path.Dir(os.Args[0])
if !path.IsAbs(os.Args[0]) {
here = path.Join(os.Getwd(), here)
}
This will get you the path to where the executable is. If you're guaranteed the user won't move this without moving the rest of your assets, you can use this, but I find it much more flexible to use the above -cwd idea, because then the user can place the executable anywhere on their system and just point it to the assets.
The best option would probably be a mixture of the two. If the user doesn't supply a -cwd flag, they probably haven't moved anything, so infer from arg 0 and the cwd. The cwd flag overrides this.

Intltool with an autoconf-generated .desktop file

In the Emperor project, I'm having some issues getting intltool to work when doing an out-of-tree build. When running make check out-of-tree, which is one of the things make distcheck does, intltool fails thus:
INTLTOOL_EXTRACT="/usr/bin/intltool-extract" XGETTEXT="/usr/bin/xgettext" srcdir=../../po /usr/bin/intltool-update --gettext-package emperor --pot
can't open ../../po/../data/emperor.desktop.in: No such file or directory at /usr/bin/intltool-extract line 212.
intltool is looking for emperor.desktop.in, which is listed in po/POTFILES.in, in the source tree. However, emperor.desktop.in is generated by the configure script from a file called emperor.desktop.in.in, in order to insert the installed executable path as configured by the user, and lands in the build tree.
These are the relevant bootstrap.sh lines:
echo +++ Running intltoolize ... &&
intltoolize --force --copy &&
cat >>po/Makefile.in.in <<EOF
../data/_column_names.h:
cd ../data && \$(MAKE) _column_names.h
EOF
The setup code in configure.ac:
IT_PROG_INTLTOOL([0.35.0])
GETTEXT_PACKAGE=emperor
AC_SUBST(GETTEXT_PACKAGE)
AC_DEFINE_UNQUOTED([GETTEXT_PACKAGE], ["$GETTEXT_PACKAGE"],
[The domain to use with gettext])
AM_GLIB_GNU_GETTEXT
data/emperor.desktop.in is listed in AC_CONFIG_FILES.
data/Makefile.am contains these lines:
desktopdir = $(datadir)/applications
desktop_in_files = emperor.desktop.in
desktop_DATA = $(desktop_in_files:.desktop.in=.desktop)
#INTLTOOL_DESKTOP_RULE#
and po/POTFILES.in contains the line
data/emperor.desktop.in
You can review all the details in the public git repository if you wish.
Can I somehow tell intltool that this file will be located in the build tree, not in the source tree? Otherwise, my options appear to be to break make distcheck (not a great option), or to ship a desktop file that doesn't include the full path and assumes that the executable is installed in the PATH. (just as messy, IMHO) - Any other options?
In your source code you have emperor.desktop.in.in, which does not seem to be in any rule as a dependency. That file has to be converted first to emperor.desktop.in and later to emperor.desktop, which does not seem to be the case in your data/Makefile.am.
desktopdir = $(datadir)/applications
desktop_in_in_files = emperor.desktop.in.in
desktop_in_files = $(desktop_in_in_files:.desktop.in.in=.desktop.in)
desktop_DATA = $(desktop_in_files:.desktop.in=.desktop)
#INTLTOOL_DESKTOP_RULE#
[...]
EXTRA_DIST = \
$(desktop_in_in_files) \
[...]
$(desktop_in_in_files) contains $(desktop_in_in_files), and Makefile will know how to deal with that.
Some further digging has brought me believe that the answer is: intltool does not support source files that aren't source files in the project. Ergo, any additional processing must be done after intltool is through
Intltool requires the lines in POTFILES to be relative to the (build-time) working directory. The file POTFILES is generated by the configure script from POTFILES.in with a simple sed script defined in the IT_PO_SUBDIR autoconf macro (called by IT_PROG_INTLTOOL) that simply prepends the relative location of the top-level source directory to the paths. Alas, modifying POTFILES does not help: the intltool-extract script does everything it can to get the source directory right. I don't believe files that are sometimes inside and sometimes outside the source tree can be supported without modifying intltool itself.

Resources