libsigrok/HACKING

256 lines
9.7 KiB
Plaintext

-------------------------------------------------------------------------------
HACKING
-------------------------------------------------------------------------------
Coding style
------------
This project is programmed using the Linux kernel coding style, see
http://lxr.linux.no/linux/Documentation/CodingStyle for details.
Please use the same style for any code contributions, thanks!
Contributions
-------------
- In order to contribute you should ideally clone the git repository and
let us know (preferably via IRC, or via the mailing list) from where to
pull/review your changes. You can use github.com, or any other public git
hosting site.
- Alternatively, patches can be sent to the development mailinglist at
sigrok-devel@lists.sourceforge.net (please subscribe to the list first).
https://lists.sourceforge.net/lists/listinfo/sigrok-devel
Adding a new hardware driver
----------------------------
The simple, scripted way (recommended):
---------------------------------------
Use the 'new-driver' script from the sigrok-util repo:
$ git clone git://sigrok.org/sigrok-util
$ cd sigrok-util/source
$ ./new-driver "Tondaj SL-814"
The example above generates a patch file against the current libsigrok
development git tree which adds a simple "stub" driver for your device
(the Tondaj SL-814 sound level meter in this case).
You can apply it like this:
$ cd libsigrok
$ git am 0001-tondaj-sl-814-Initial-driver-skeleton.patch
You can now edit the files in src/hardware/tondaj-sl-814 as needed
and implement your driver based on the skeleton files there. That means your
patch submission later will consist of at least two patches: the initial one
adding the skeleton driver, and one or more additional patches that actually
implement the respective driver code.
The manual way:
---------------
This is a rough overview of what you need to do in order to add a new driver
(using the Tondaj SL-814 device as example). It's basically what the
'new-driver' script (see above) does for you:
- Makefile.am: Add HW_TONDAJ_SL_814 and add to libsigrok_la_SOURCES.
- configure.ac: Add a DRIVER() and DRIVER2() call.
- src/drivers.c: Add a tondaj_sl_814_driver_info entry in two places.
- src/hardware/tondaj-sl-814/ directory: Add api.c, protocol.c, protocol.h.
See existing drivers or the 'new-driver' output for the details.
Random notes
------------
- Don't do variable declarations in compound statements, only at the
beginning of a function.
- Generally avoid assigning values to variables at declaration time,
especially so for complex and/or run-time dependent values.
- Consistently use g_*malloc() / g_*malloc0(). Do not use standard
malloc()/calloc() if it can be avoided (sometimes other libs such
as libftdi can return malloc()'d memory, for example).
- Always properly match allocations with the proper *free() functions. If
glib's g_*malloc()/g_*malloc0() was used, use g_free() to free the
memory. Otherwise use standard free(). Never use the wrong function!
- We assume that "small" memory allocations (< 1MB) will always succeed.
Thus, it's fine to use g_malloc() or g_malloc0() for allocations of
simple/small structs and such (instead of using g_try_malloc()), and
there's no need to check the return value.
Do use g_try_malloc() or g_try_malloc0() for large (>= 1MB) allocations
and check the return value.
- You should never print any messages (neither to stdout nor stderr nor
elsewhere) "manually" via e.g. printf() or g_log() or similar functions.
Only sr_err()/sr_warn()/sr_info()/sr_dbg()/sr_spew() should be used.
- Use glib's gboolean / TRUE / FALSE for boolean types consistently.
Do not use <stdbool.h> and its true / false, and do not invent private
definitions for this either.
- Consistently use the same naming convention for #include guards in headers:
<PROJECTNAME>_<PATH_TO_FILE>_<FILE>
This ensures that all #include guards are always unique and consistent.
Example: LIBSIGROK_HARDWARE_MIC_985XX_PROTOCOL_H
- Consistently use the same naming convention for API functions:
<libprefix>_<groupname>_<action>().
Examples:
sr_log_loglevel_set(), sr_log_loglevel_get(), sr_log_handler_set(),
sr_log_handler_set_default(), and so on.
Or:
sr_session_new(), sr_session_destroy(), sr_session_load(), and so on.
Getter/setter function names should usually end with "_get" or "_set".
Functions creating new "objects" should end with "_new".
Functions destroying "objects" should end with "_destroy".
Functions adding or removing items (e.g. from lists) should end with
either "_add" or "_remove".
Functions operating on all items from a list (not on only one of them),
should end with "_all", e.g. "_remove_all", "_get_all", and so on.
Use "_remove_all" in favor of "_clear" for consistency.
- All enums should generally use an explicit start number of 10000.
If there are multiple "categories" in the enum entries, each category
should be 10000 entries apart from the next one. The start of categories
are thus 10000, 20000, 30000, and so on.
Adding items to an enum MUST always append to a "category", never add
items in the middle of a category. The order of items MUST NOT be changed.
Any of the above would break the ABI.
The enum item 0 is special and is used as terminator in some lists, thus
enums should not use this for "valid" entries (and start at 10000 instead).
Doxygen
-------
- Use the @ notation for all Doxygen comments (e.g. @param, not \param).
- Do not use the @brief tag, it's unnecessary as we use JAVADOC_AUTOBRIEF.
- Generally use the following item order in Doxygen comments:
- Brief function description (1 line), followed by an empty line.
- Optionally, a longer function description (and another empty line).
- The list of parameter descriptions (@param).
- The return value description (@return or @retval).
- An optional @since tag (only for public SR_API functions).
- An optional @private tag (for private SR_PRIV functions).
- In @param lines, the name of the parameter is followed by a space and
then a sentence describing the parameter (starts with a capital letter,
ends with a full stop).
- In Doxygen comments, put an empty line between the block of @param lines
and the final @return line. The @param lines themselves (if there is more
than one) are not separated by empty lines.
- Mark private functions (SR_PRIV) with /** @private */, so that Doxygen
doesn't include them in the output. Functions that are "static" anyway
don't need to be marked like this.
- Mark private variables/#defines with /** @cond PRIVATE */ and
/** @endcond */, so that Doxygen doesn't include them in the output.
Variables that are "static" don't need to be marked like this.
- Mark all public API functions (SR_API) with a @since tag which indicates
in which release the respective function was added (e.g. "@since 0.1.0").
If the function has existed before, but its API changed later, the @since
tag should mention only the release when the API last changed.
Example: The sr_foo() call was added in 0.1.0, but the API changed in
the later 0.2.0 release. The docs should read "@since 0.2.0" in that case.
Non-public functions (static ones, and those marked SR_PRIV) don't need
to have @since markers.
The @since tag should be the last one, i.e. it should come after @param,
@return, @see, and so on.
- Examples:
/**
* Tell a hardware driver to scan for devices.
*
* In addition to the detection, the devices that are found are also
* initialized automatically. On some devices, this involves a firmware upload,
* or other such measures.
*
* The order in which the system is scanned for devices is not specified. The
* caller should not assume or rely on any specific order.
*
* Before calling sr_driver_scan(), the user must have previously initialized
* the driver by calling sr_driver_init().
*
* @param[in] driver The driver that should scan. Must be a pointer to one of
* the entries returned by sr_driver_list(). Must not be NULL.
* @param[in] options List of 'struct sr_hwopt' options to pass to the driver's
* scanner. Can be NULL/empty.
*
* @return A GSList * of 'struct sr_dev_inst', or NULL if no devices were
* found (or errors were encountered). This list must be freed by the
* caller using g_slist_free(), but without freeing the data pointed
* to in the list.
*
* @since 0.2.0
*/
/**
* Query value of a configuration key at the given driver or device instance.
*
* @param[in] driver The sr_dev_driver struct to query. Must not be NULL.
* @param[in] sdi (optional) If the key is specific to a device, this must
* contain a pointer to the struct sr_dev_inst to be checked.
* Otherwise it must be NULL. If sdi is != NULL, sdi->priv must
* also be != NULL.
* @param[in,out] data Pointer to a GVariant where the value will be stored.
* Must not be NULL. The caller is given ownership of the GVariant
* and must thus decrease the refcount after use. However if
* this function returns an error code, the field should be
* considered unused, and should not be unreferenced.
*
* @retval SR_OK Success.
* @retval SR_ERR Error.
* @retval SR_ERR_ARG The driver doesn't know that key, but this is not to be
* interpreted as an error by the caller; merely as an indication
* that it's not applicable.
*
* @since 0.3.0
* @private
*/
Testsuite
---------
You can run the libsigrok testsuite using:
$ make check
Release engineering
-------------------
See
http://sigrok.org/wiki/Developers/Release_process
for a list of items that need to be done when releasing a new tarball.