The segfault can be triggered by using any diskio object with a
non-existent device, as prepare_diskio_stat() then returns 0 and the
call to obj->data.diskio->current in conky.c:4050 pulls the trigger.
In fact, it's not a problem when the device doesn't exist, as
update_diskio() simply won't fill in any values. So skip the check and
upon device node appearance everything goes it's normal way.
While there, also eliminate double readout of the last line of
/proc/diskstats: after the last line has been read, FEOF is not yet set.
BUT fgets() will return NULL when trying to read the next line. So
better check for fgets()'s return value instead of using feof().
Also strncmp() is useless here, since we really want to compare the full
paths. Besides, text_buffer_size also should be big enough to not make a
difference here.
Also, note that free() does not zero the passed pointer. So conditional
freeing always needs to look like this:
| if (var) {
| free(var);
| var = 0;
| }
* change the height of execbar and execibar to be the same as other
bars have by default (6 pixels)
* treat mixer values as percentages as they're usually 0-100 (switch
to spaced_print() with pad_percents)
* change temp_print() from snprintf() to spaced_print()
* remove decimals from temp_print() as none of the current sources can
supply values smaller than 1 degree (C or F, doesn't matter)
* add a space between number and the unit in human_readable()
* fix number printing in human_readable()
* network $upspeed and $downspeed now use human_readable()
This makes it possible for the compiler to do better optimizations by
adding "static" to functions which do not need to be exported.
Since some of them (e.g. the compare_*() functions) are called very
often, this may decrease conky's need of resources a bit.
Since there is no choice of which moc player to get information from,
all moc objects' data source can be identical. Also hide some internal
data (the thread e.g.). Since from now on there can only be one moc
thread (not highlander ;), we don't need to treat the thread object
specially.
While here, fix indenting.
All objects parsing text into objects formerly parsed and evaluated
their arguments each update interval. This does only make sense when the
parsed text could change between updates, which is the case for execp
and execpi objects, but none of the others. So have them parse the text
when creating them, so each update interval only the output has to be
re-extracted.
In fact, this should give a performance boost when using any of the
affected objects, as parsing the input text is quite expensive since
we're evaluating templates.
Using a global ifblock stack for all parsed ifblock objects causes
problems when doing sub-parsing in objects taking other objects as
parameters, because the possibly non-empty stack at startup leads to
false alarm when checking for stack emptiness after parsing the objects.
Use a void ** as the object to pass around, so callers don't need to
know struct ifblock_stack_obj.
In fact, they are. But we don't want to treat them as those, because
the rewrite made the only situation when threaded objects matter
uncritical, i.e. when an object calls calling parse_conky_vars().
In general, argument types should match, but for combined long and
double usage the long is being converted to double before evaluation.
A few examples:
${if_match ${cpu} < 30}
${if_match "asdf" != "qwer"}
${if_match 0.5 < 0.50001}
${if_match 49.999 < 50}
The crux is to split args between those for scan_graph() on one hand and
prepare_diskio_stat() on the other hand. To make the code working with
minimal changes, move the yet optional devicename to the end.
General note on graphs:
- for all graphs there exist the optional arguments for scan_graph()
which come first and are optional (intended duplicate)
- all other args are object specific, and may be optional as well
Handle mpd internal information inside mpd.c. Use a refcounter to check
if the mpd-information can be freed (maybe useless). Remove the now
useless "full" flag of free_text_objects.
This was a bug, not a feature. From the FreeBSD manpage rtentry(9):
| Vt struct sockaddr *rt_gateway;
|
| The "target" of the route, which can either represent a destination in
| its own right (some protocols will put a link-layer address here), or
| some intermediate stop on the way to that destination (if the
| RTF_GATEWAY flag is set).
So for a default route with "dest" and "mask" both zero, either
RTF_GATEWAY is present, or "gate" is zero.
The only change of behaviour occurs when adjusting the unit of positive
values. For some reason 1000LL was used, which has now been replaced by
1024LL. Usage of abs() and MAX() might be OS dependent, but it should
not be too hard to implement them by hand if they're missing somewhere.
It was a static int before, holding the (already defined) enum's value,
which doesn't make sense. Also it's used only inside conky.c. Since
enums can't be assigned other values than what are defined, we can skip
checking for illegal spacer value at each invocation of spaced_print,
and then also drop the function name argument, which apparently didn't
make sense at all, because use_spacer is a global option.
Instead of splitting information, use diskio_stats[0] in diskio.c for
the totals. This saves a few branches, and frees some data from struct
information.
Some statics are now defined by configure.ac.in, conky.h got a lot
smaller, which I see as a positive sign.
This patch should not change any functionality, just change what's
defined where. Some features I did/could not test are probably broken,
also compiling for any other OS surely won't complete now. Though I
think fixing these problems is as easy as including some missing
headers.
I've done compile checks with the following configure options:
./configure --enable-eve --enable-rss --enable-smapi
--enable-wlan --enable-debug --enable-testing
So what needs to be tested is:
- audacious
- BMPx
- xmms2
- nvidia
Using a macro is not possible anymore, so have a function dev_name() do
the job. Functionality is the same as the old DEV_NAME, unless the
specified path is a symlink in which case it's being dereferenced first.
To save callers from having to free the returned string, a static local
buffer is used (effectively rendering the function non-reentrant).
This patch makes Conky print a sample config when being called with the
'-C' flag.
A short test showed an increase of ~10kbytes of the conky binary's size.
From strdup(3):
| If s is longer than n, only n characters are copied, and a
| terminating null byte ('\0') is added.
So allocate at most n+1 bytes and make sure the last one is zero, as
strncpy() doesn't add it itself.
So in fact to allow a maximum space for string dup of 23, strndup() has
to be called like this:
| dup = strndup(src, 23 - 1);
FIXME: Find the critical points in code this change touches and make
sure the invocation there is correct.
Two things that pissed me off about the old one:
* only limited support for nesting templates
* totally broken output when using conditionals inside a template
The later one was the hard one to fix. ;)
It requires to already have the full text substituted before the text
objects are being created from it. Generating only the contained objects
broke, because the conditionals got wrong offsets to jump to.
After that was fixed, full nesting support is realised by simply
repeating the replacement until no more template objects are found.
Withdrawn windows are those you get from applications supporting the
'-w' flag, like e.g. gkrellm or all those sweet WindowMaker dockapps
(wmcpu and Co.). In Fluxbox, these windows are drawn into the slit. Most
other window managers put them into their "taskbar", AFAIK.
This patch makes Conky act exactly the same if the following settings
are selected:
| own_window yes
| own_window_type dock
This introduces a new configuration variable called "temperature_unit",
specifying the unit of all temperature sensors. To achieve this, each
object outputting a temperature has to call temp_print() like so:
| temp_print(p, p_max_size, <temp val as double>, <unit of val>);
to specify the input temperature unit, either one of the constants
TEMP_CELSIUS or TEMP_FAHRENHEIT.
The new command line option '-D' ('--debug') increases debugging level by one.
For debugging output a user could be interested in, use the macros DEBUG() and
DEBUG2(). Functionality is equal to the ERR() macro. DEBUG2() prints stuff only
if debugging level is greater one, which means that '--debug' has been
specified more than once. This patch also includes usage of the macros for the
new template object (as debugging syntax errors in templates is one thing a
user potentially needs to do).
git-svn-id: https://conky.svn.sourceforge.net/svnroot/conky/trunk/conky1@1273 7f574dfc-610e-0410-a909-a81674777703
* handle escaped characters in parameters to the template object, too
* add samples to the documentation
* do some code cleanup by outsourcing the string substitution into it's
own function
git-svn-id: https://conky.svn.sourceforge.net/svnroot/conky/trunk/conky1@1271 7f574dfc-610e-0410-a909-a81674777703
Basically you can define a template like this:
| template0 \1:\n ${fs_free \2} / ${fs_used \2}
and use it later like this:
| ${template0 root /}
| ${template0 boot /boot}
which is exactly the same as the following:
| root:
| ${fs_free /} / ${fs_used /}
| boot:
| ${fs_free /boot} / ${fs_used /boot}
hope you like it, I do already. :)
git-svn-id: https://conky.svn.sourceforge.net/svnroot/conky/trunk/conky1@1270 7f574dfc-610e-0410-a909-a81674777703
* happened when passing wrong arguments to ${scroll}, particularly with only
one instead of the mandatory two unsigned ints
* as it's not clear whether %n will change the value returned by sscanf or not,
use a solution that works in both situations (%n incrementing the return
value or not)
git-svn-id: https://conky.svn.sourceforge.net/svnroot/conky/trunk/conky1@1268 7f574dfc-610e-0410-a909-a81674777703
