build: check for unexpanded macros in configure.ac
We're now using m4 macros from the autoconf-archive collection. The
user may be missing this package so check for unexpanded AX_* macros
and bail-out if any are found.
core: events: correctly handle POLLNVAL in gpiod_line_event_wait()
The only error that can be indicated by ppoll() in the revents field
of struct pollfd when polling a line events file descriptor is
POLLNVAL. It can happen if the user calls close() on said descriptor
after retrieving it with gpiod_line_event_get_fd().
Currently we would act as if there's a line event to read. Make
gpiod_line_event_wait() and its bulk variant return -1 and set errno
to EINVAL in this case.
The copyright notice displayed together with the package version using
the --version argument got accidentally modified by commit 0704bcdbbf56
("all: update the copyright notice") and is now incorrect. Fix it.
core: use reference counting for line file descriptors
In v0.x series we were using a separate, refcounted structure for
storing the file descriptor returned by the linehandle request ioctl().
We also had a separate mechanism for event file descriptors. This was
quite complicated and was replaced by a simpler structure in v1.x
series.
It turned out however that we're now calling close() on already closed
descriptors when releasing a set of simultaneously requested lines.
In order to fix that we need to go back to having a refcounted file
descriptor handle. This time we're using the same structure for storing
both values and events file descriptors (even though there can't be
more than one line per descriptor requested for events) for simplicity.
This patch also adds a test case that verifies the reference counting
works as expected.
build: use separate versioning schemes for API & ABI
Commit 783ff2e3c707 ("API: start certain enums from 1") broke the ABI
compatibility (while keeping the source compatibility). We need to
indicate that to the users by changing the SONAME string. Up until now
the SONAME would correspond with the major release number. We now need
to have a separate versioning schemes for API and ABI.
Keep using the same MAJOR.MINOR.RELEASE scheme for the API, but switch
to using the libtool versioning - CURRENT.REVISION.AGE - for ABI. Start
the ABI version from 2.0.0 to indicate that it changed since 1.0.0.
Define a separate ABI version for the C++ bindings, but start with 0.0.0
since it was not a part of any release yet.
The most commonly used cleanup attribute is the one used for automatic
closing of GPIO chips. For brevity: hide the long TEST_CLEANUP(...)
invocation behind a shorter TEST_CLEANUP_CHIP macro.
Clemens Gruber [Tue, 20 Feb 2018 16:57:32 +0000 (17:57 +0100)]
tools: move function attributes to the header file
The attributes must be located next to function declarations in the
header file. Otherwise, the NORETURN attribute seems to have no effect
and GCC emits several Wimplicit-fallthrough warnings when compiling
libgpiod. This patch fixes those warnings.
Signed-off-by: Clemens Gruber <clemens.gruber@pqgruber.com> Signed-off-by: Bartosz Golaszewski <bartekgola@gmail.com>
For enums describing direction & active state, event types etc. use 1
as the first value as it's easier to debug when all valid values are
different than in a zero-initialized data structure.
Also: the master branch doesn't contain the tags for bugfix releases so
documenting them in this branche's NEWS makes no sense.
Fold all the info about bugfix releases into the next minor release and
only keep major and minor releases in NEWS on the master branch. Bugfix
releases will only be documented in their respective minor branches
from now on.
doc: improve the doxygen description of default_val(s)
Users report being confused by the name 'default_val(s)' for function
parameters. Because it's in line with the kernel uAPI, we'll leave it
like this, but let's improve the documentation a bit by stating
explicitly that this parameter represents the initial values for
requested GPIO lines.
API: use gpiod_ctxless_ as prefix for the high-level API
As suggested by Linus Walleij: the word 'simple' is generally
subjective and the high-level routines provided by libgpiod are not
necessarily simple, but rather contextless, as they don't require any
resource managemend. Use 'gpiod_ctxless_' as prefix for all high-level
symbols.
Clemens Gruber [Mon, 22 Jan 2018 16:12:13 +0000 (17:12 +0100)]
gpiomon: initialize the active low variable
The variable active_low is not initialized, so the value is undefined
and if the -l argument is not passed, the undefined value remains. Fix
it by initializing it to false.
Signed-off-by: Clemens Gruber <clemens.gruber@pqgruber.com> Signed-off-by: Bartosz Golaszewski <bartekgola@gmail.com>
This project contains both a library and executables so it's better to
use an umbrella term "This program ..." than "This library ..." in all
the source files.
core: don't allow open-source and open-drain flags at the same time
If the hardware actually allowed it, the electrical result would be
disastrous. This has only recently been fixed in the kernel, so add
a relevant check to the library as well.
We don't need to release lines manually - line resources are managed
by GPIO chip objects. Remove the calls to release functions wherever
they don't make sense for any subsequent test code.
Rework the way we verify that line bulk objects are consistent: split
the verify_line_bulk() function into sub-routines, make the names more
self-explanatory and add other minor fixes & tweaks.
Improve the consistency of errno codes: set errno to EPERM if the line
for which gpiod_line_event_get_fd() is called is not configured for
event monitoring.
We were already testing that line requests fail if we pass them a line
bulk containing lines from different chips, but add an analogous test
for gpiod_line_get_value_bulk().
We plan to change the way gpio-mockup platform devices are registered
in the kernel. This change will add an ID to the gpio-mockup platform
device name and will result in the devpath being changed to
'/devices/platform/gpio-mockup.X/gpiochip'.
When checking the devpath during the test case setup, it's enough to
verify that we got an event from the correct platform device - we
don't need to check the gpiochip name. Shorten the string against
which we check to skip whatever comes after the platform device name.
While we're at it: move the string definition into the function which
is the only user.
iter: fix a crash occurring if no GPIO chips are present
If no GPIO chips are present in the system, we bail out from
gpiod_chip_iter_new() before allocating the chips array. Add a check
in gpiod_chip_iter_free() to avoid calling free() for iter->chips in
that case.
The changes roughly correspond with what we did for chip iterators.
Only perform operations that can fail in gpiod_line_iter_new(), so
that there's no need to check for errors after every call to
gpiod_line_iter_next().
Make gpiod_line_iter an opaque pointer and provide 'new' and 'free'
functions.
Update all users and relevant test cases. Extend the test API with
a new cleanup function.
The 'next' routines for chip iterators can no longer fail. Update
the docs accordingly. Also: turn the extended description into a note
for the noclose variant.
Many test cases use asprintf() to build custom strings. This routine
can fail and the return value is tested using the TEST_ASSERT macros
but this is not the subject of the actual testing. The test case must
also free the string allocated by this function.
As a simplification: pull this functionality into the testing
framework so that we can build custom strings without keeping track of
resources. The new function will also make the whole test suite die if
the internal asprintf() call fails.
We can significantly simplify the chip iterators by performing all
actions that can yield an error in gpiod_chip_iter_new(). Instead of
opening each subsequent chip in gpiod_chip_iter_next(), open them all
in the new function and make next just return an active handle.
Fix relevant test cases and macros, remove related routines that are
no longer needed.
tests: gpioinfo: don't use static names for GPIO chips
We need to construct the pattern strings using the names supplied by
the testing framework - otherwise we may end up getting false
negatives from these test cases if the chip layout in the system is
different than expected.
Using scandir() instead of opendir() allows us to scan for all
gpiochip devices already when creating the iterator. This simplifies
the code and doesn't require keeping the directory open when
iterating.
Storing the array of all found GPIO chips will also make it easier if
we ever want to implement a reverse iterator (e.g. for C++ bindings).
core: code shrink for name, label & consumer strings handling
Create a helper function that implements a uniform way of copying the
info strings from the kernel data structures to gpiod_line and
gpiod_chip objects.
The kernel uapi line handle and event request structures are only
needed for relevant ioctl() calls. After that they can be discarded as
all we need is the file descriptor.
Setting the open-drain or open-source flags for input doesn't make any
sense as these flags are only valid for output mode. This should
actually be checked in the kernel. For now fix the test case.
projects / qemu-gpiodev / libgpiod.git / log