Only evaluate these shell commands once at Makefile startup, otherwise
every single call to the compiler/linker will re-evaluate these lines
again and again. This speeds up the build quite a bit for me.
- Make the gkrellm.pc install target depend on its input file, fixes
install target with parallel make
- Use a single rule for writing gkrellm.pc. Having a separate rule for
windows breaks parallel builds, for some reason a parallel build of just
the install_windows target ends up having unix gkrellm.pc contents.
Avoid this possible build race by using a single rule with optional
contents at the end of the file.
- Enforce static linking against libgkrellm.a on win32, otherwise plugin
sources using autotools/libtool fail to link. libtool tries very hard
to avoid mixing static and shared libraries so we will have to fall back
to raw linker flags (libgkrellm.a on win32 is kind of a static
convenience library)
- Handle Windows 6.1 - 6.3 which equals Windows 7, 8 and 8.1 including
its server variants
- Shorten the client/server differentiation by using a variable
The calling GKrellM code always expects a system name that can be split
at the first space character to provide kernel name and kernel version.
Return a string containing a space to avoid a crash on any Windows
version the code does not know about (or in case querying the version
failed).
Recent Gtk+ versions seem to defer creation of native windows even
further which made setting HWND_TOPMOST a no-op. Replace this lowlevel
hack with the standard Gtk+ way of making a window always-on-top, it
now works as expected.
A patch from Joe Garcia. A ssh tunnel can be up but possibly not
connected to anything in which case gkrellm_getline() can return 0.
This patch prevents a busy loop by limiting the 0 return retries.
Edit one of the CPU chart text format string options.
Fix gkrellm_category_vbox() so the returned vbox will be expanded like
it should be. This exposed a clock config box packing needing a change.
The Disk Launch Commands vbox was packed into the wrong box.
- Replace SMP real CPU group enable with individual real CPU enables.
I've had requests for this feature from people with 24 - 32 cpu machines
where there is insufficient vertical space for all cpu charts and they
want to split the CPU charts into multiple gkrellm stacks with --config.
- Add option to show/hide CPU panel labels so only a krell can be in the
panels. Not showing CPU panel labels saves vertical space - for systems
with medium to large numbers of CPUs. I find it useful with 4+ CPUs.
To make up for no panel label, added new chart label format string options
that draw the CPU label on the chart.
Temp and fan sensors cannot be relocated to CPU panels if labels not shown.
mail.c: replace deprecated g_mutex_new() with g_mutex_init() and
fix const discarded warnings.
plugins.c: gkrellm_place_plugin() change code that inserts plugins into
monitor_list to avoid compiler warnings.
Patch from Ville Skytta: fix issues flagged by cppcheck:
- netbsd.c: add some close(fd) calls before returns in gkrellm_sys_battery_init().
- winops-win32.c: add a free() in gkrellm_winop_place_gkrellm().
Bugfixes:
- chart.c: flag config modified when changing a chart's Resolution per Grid.
- fix memory leaks in multiple files: add g_thread_unref() for all GThread
pointers returned from g_thread_new() calls.
Parsing an IPv6 prefix length has to reset errno prior to calling
strtoul() to reliably detect errors (the return value is not suitable
for detecting errors).
Patch taken straight from Debian BTS #764023. Thanks for Peter Denison
for spotting this and providing a patch.
Always enable ASLR and DEP when building GKrellM on Windows. Since
GKrellM also does network communication bugs like buffer overflows and
the like may eventually allow remote code execution. By enabling these
build flags such attacks should be quite a bit more complicated
(although still not impossible).
Add a dependency on the windows build target to allow building and
installing GKrellM using a single target. This fixes calling "make
install_windows" without a prior call to "make windows".
- Export the CC and PKG_CONFIG variables before calling configure scripts,
especially for cross compilation and modified environments, otherwise
configure might detect libraries that are not usable by the compiler
later on
- Replace the hardcoded pkg-config calls in configure with a variable
that may be overridden by Makefile
- Make the server configure script log the environment just like the
configure script used for the client.
- Honour the CFLAGS env var too since the other configure does the same
Do not try to execute generated executables. Altough this is a nice
idea in theory it breaks on cross compile environments which usually
cannot execute binaries of the target platform.
Remove Gtk+2 compile and link flags from all configure tests because:
- no test code depends on Gtk+2
- gkrellmd does not need Gtk+2 at all
- the cflags variable in the src/configure was misspelled and thus did
not have any effect
Override all tools with prefixed versions if the CROSS variable has
been set. This is the recommended way to enable cross compilation using
mxe without touching lots of environment variables first.
Move all make variables and rules duplicated between gkrellm and
gkrellmd makefiles into a common file that can be included. This
avoids duplication and hopefully avoids the two makefiles getting out
of sync too much. This is also a preparation to ease overriding
toolchain binaries (compiler, linker, pkgconfig) as part of windows
cross compiling.
This change also puts the "ar" tool into a variable that can be
overridden if needed.
Please note that this change uses the GNU Make specific include
statement but AFAICS the makefiles already depend on GNU extensions in
other areas (":=" and "ifdef" are non-standard make syntax).
Apply a slightly modified version of a patch provided by Pierre Neyron
that allows to enable/disable new network interfaces by default.
This feature is especially helpful for VM or container setups where many
network interfaces are created.
Adding md5c to the build is already handled by src/Makefile but that
one only kicks in if EXTRAOBJS is not set at all.
This fixes accidental build (and inclusion) of the md5 code that is
normally used from OpenSSL or GnuTLS if one of them is detected.
GLib version 1 is very outdated by now and not available in any recent
Linux distribution. Keeping this code around without actually being
able to test it makes little sense so let's get rid of that.
- Always check the return value of
g_win32_get_package_installation_directory_of_module, this call may
fail and return NULL
- g_build_filename is documented as never returning NULL, no need to
check for that
- Use g_thread_new instead of g_thread create
- Drop explicit call of g_thread_init, it is done implicitely by GLib
since 2.32
- Since this bumps the minimum required GLib version add and extend the
pkg-config based version checks to error out on older GLib versions
The remaining codebase uses g_strdup() already so do the same for the
chime command too. This fixes a gcc warning caused by a missing include
for strdup().
Move all variable and struct init code into if 0 blocks like the code
that is supposed to use this. Eventually this whole cruft can be
dropped altogether in case GKrellM can be ported to use Gtk+ session
management.
- Adding a sensor in the mbmon support stored the added sensor in a
pointer but never used it afterwards
- Drop pixbuf width in draw_left_frame_overlap, the code unlike other
drawing helpers does not use the pixbuf width
- Remove unused scroll selection from net_stats_page
This fixes several set but unread warnings when compiling with a recent
gcc (encountered with gcc 4.8.2).
Read sensor information from the GPU-Z shared memory area if present.
As with all sensor backends on Windows this needs the GPU-Z tool
running in the background.
Do not use a fixed 5 second delay after having a started a windows
sensor application, instead try to access the shared memory area
repeatedly and only give up after 10 seconds. Please note that for this
feature to be usable the sensor app must be found in PATH.
- Move opening and closing of a win32 shared memory area into helper
functions. shm_open ensures that the data pointer is valid and
otherwise gets rid of the opened file handle right away. shm_close
cleans up both the handle and the data pointer if needed.
- Redo the initial check for presence of a sensor tool that uses a
shared memory area for its sensor data by using the above two
functions
- Port all sensor value reading to use shm_open/shm_close as well
This gets rid of some code duplication and also lowers the indentation
level in several areas which hopefully makes the code a bit easier to
read. Additionally all the win32 API calls are limited to two functions.
Stefan Gehn
Bill Wilson
Jindřich Makovička
Ville Skyttä