Some drivers set the device instance list to NULL in their scan() callback.
This means the driver loses all references to any devices contained in that
list and their resources will be leaked. Drivers can't free the devices at
this point either since an application might still use a device on the
list. So the existing devices on the instance list need to remain
unmodified during the scan() callback, even if that means that there will
be duplicates on the instance list. Only an explicit invocation of
sr_dev_clear() by the application is allowed to free the devices on the
instance list and reset the list.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
The sigrok core needs a list of all available drivers. Currently this list
is manually maintained by updating a global list whenever a driver is added
or removed.
Introduce a new special section that contains the list of all drivers. The
SR_REGISTER_DEV_DRIVER() and SR_REGISTER_DEV_DRIVER_LIST() macro is used to
add drivers to this new list. This is done by placing the pointers to the
driver into a special section. Since nothing else is in this section it is
known that it is simply a list of driver pointers and the core can iterate
over it as if it was an array.
The advantage of this approach is that the code necessary to add a driver
to the list is completely contained to the driver source and it is no
longer necessary to maintain a global list. If a driver is built it will
automatically appear in the list, if it is not built in won't. This means
that the list is always correct, whereas the previous approach used ifdefs
in the global driver list file which could get out-of-sync with the actual
condition when the driver was built.
Any sr_dev_driver structs that are no longer used outside the driver module
are marked as static.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Most drivers have a forward declaration to their sr_dev_driver struct at
the beginning of the driver file. This is due to historic reasons and often
no longer required. So remove all the unnecessary forward declarations.
Some drivers still require the forward declaration, but only reference the
driver struct from within the driver scan() callback. Since the driver
struct is passed to the scan callback replace the references to the global
variable with the local parameter. In some cases this requires adding the
parameter to some of the helper functions that are called from the scan()
callback.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Now that the signature of std_init() matches that of the driver init()
callback we can remove all wrapper functions around std_init() and use it
directly as the init() callback.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
std_init() checks if the pass in struct sr_dev_driver is non-NULL and
prints a error message and returns an error if it is NULL.
std_init() is exclusively called from driver init() callbacks for which the
core already checks if the struct sr_dev_driver is non-NULL before invoking
the callback. This means the check in std_init() will always evaluate to
false. So drop this check.
This also means that the prefix parameter that was used in the error
message is no longer needed and can be removed from the function signature.
Doing so will make the std_init() function signature identical to the
init() callback signature which will allow to directly use it as such.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
The std_init() callback has the order of the first two paramters opposite
to the init() callback. This is primarily due to historical development.
Since the std_init() function is usually called from a driver's init()
callback aligning the order will allow direct register pass through rather
than having to swap them around. It also allow to eventually use the
std_init() function directly as the init() callback.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Every single hardware driver has the very same implementation of the
dev_list() callback. Put this into a helper function in the standard helper
library and use it throughout the drivers. This reduces boiler-plate code
by quite a bit.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
std_init() allocates a drv_context struct which needs to be freed by the
driver in its cleanup struct. But the vast majority of drivers does never
does this causing memory leaks.
Instead of addressing the issue by manually adding code to free the struct
to each driver introduce a new helper function std_cleanup() that takes
care of this. In addition to freeing the drv_context struct std_cleanup()
also invokes sr_dev_clear() which takes care of freeing all devices
attached to the driver.
Combining both operations in the same helper function allows to use
std_cleanup() as the cleanup callback for all existing drivers, which
reduces the amount of boiler-plate code quite a bit.
All drivers are updated to use the new helper function.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
This fixes parts of bug #423.
The list of fixed warnings:
src/output/srzip.c:285:3: warning: Value stored to 'ret' is never read
ret = zip_append(o, logic->data, logic->unitsize, logic->length);
^ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
src/scpi/scpi.c:610:2: warning: Value stored to 'ret' is never read
ret = SR_OK;
^ ~~~~~
src/scpi/scpi.c:667:2: warning: Value stored to 'ret' is never read
ret = SR_OK;
^ ~~~~~
src/dmm/vc870.c:410:2: warning: Value stored to 'info_local' is never read
info_local = (struct vc870_info *)info;
^ ~~~~~~~~~~~~~~~~~~~~~~~~~
src/hardware/conrad-digi-35-cpu/api.c:130:2: warning: Value stored to 'ret' is never read
ret = SR_OK;
^ ~~~~~
src/hardware/fx2lafw/api.c:658:2: warning: Value stored to 'timeout' is never read
timeout = fx2lafw_get_timeout(devc);
^ ~~~~~~~~~~~~~~~~~~~~~~~~~
src/hardware/gmc-mh-1x-2x/protocol.c:941:3: warning: Value stored to 'retc' is never read
retc = SR_ERR_ARG;
^ ~~~~~~~~~~
src/hardware/gmc-mh-1x-2x/api.c:168:2: warning: Value stored to 'model' is never read
model = METRAHIT_NONE;
^ ~~~~~~~~~~~~~
src/hardware/ikalogic-scanalogic2/api.c:325:2: warning: Value stored to 'ret' is never read
ret = SR_OK;
^ ~~~~~
src/hardware/openbench-logic-sniffer/api.c:185:3: warning: Value stored to 'devc' is never read
devc = sdi->priv;
^ ~~~~~~~~~
src/hardware/rigol-ds/api.c:813:3: warning: Value stored to 'devc' is never read
devc = sdi->priv;
^ ~~~~~~~~~
src/hardware/scpi-pps/api.c:405:2: warning: Value stored to 'ret' is never read
ret = SR_OK;
^ ~~~~~
src/hardware/yokogawa-dlm/api.c:239:2: warning: Value stored to 'ret' is never read
ret = SR_ERR_NA;
^ ~~~~~~~~~
Since Autoconf places some important feature flags only into the
configuration header, it is necessary to include it globally to
guarantee a consistent build.
A few of these were pretty serious, like missing arguments,
passing integers where a string was expected, and so on.
In some places, change the types used by the code rather than
just the format strings.
Move the include flags for files in the source tree from
configure.ac to Makefile.am where they belong. Also use
AM_CPPFLAGS instead of CFLAGS/CXXFLAGS to make sure the
files in the build/source tree are always picked up first.
Also, remove the include/libsigrok sub-directory from the
search path, thereby making the <libsigrok/> prefix mandatory
when building libsigrok itself. This matches the convention
already imposed on users of the library.
Use g_malloc0() for small allocations and assume they always
succeed. Simplify error handling in a few places accordingly.
Don't always sanity-check parameters for non-public (SR_PRIV)
functions, we require the developers to invoke them correctly.
This allows further error handling simplifications.
Set this new parameter to 0 (no timeout) at every call site. This is
consistent with previous behaviour, so cannot cause any regressions.
Waiting forever for a serial operation is clearly always wrong. Without
specific knowledge of each device and driver however, I can't choose
appropriate timeouts for each call. The maintainers of these drivers
will need to do so, and also add appropriate handling of timeouts.
When this commit is merged, a bug should be entered for each driver
that is touched by it.
These calls are executed from an event handler and were previously nonblocking,
but have no partial write handling. They send short packets so should be OK to
block, most likely the output buffer will be empty anyway.
Fix error handling for some: serial_write can return any negative error code.
This call was previously nonblocking, but there is no handling of partial
writes. It is called from config_set where it is free to block.
Also fix error handling: serial_write can return any negative error code,
not just -1.
These calls were already nonblocking since the driver opens the port with the
SERIAL_NONBLOCK flag. They only read one byte. A return value of zero is not
handled, but should not occur in theory due to the G_IO_IN check. It might be
good to add handling of a zero return anyway, since I'm not sure if this is
always accurate.
This call was already nonblocking since the driver opens the port with the
SERIAL_NONBLOCK flag. It only reads one byte, and a zero result is handled
appropriately.
Lars-Peter Clausen
Uwe Hermann
Daniel Elstner
Soeren Apel
Martin Ling
Bert Vermeulen