Roman Joost [Mon, 22 Sep 2014 23:36:05 +0000 (09:36 +1000)]
Use `env` to start tclsh instead of /bin/bash
This patch replaces the shebang line using /bin/bash with /usr/bin/env.
The previous version obfuscates which shell is being used by the tcl
script. Any benefits of using Bash as a start up are provided by `env`.
Martin Sperl [Tue, 9 Sep 2014 10:13:45 +0000 (03:13 -0700)]
fixed argument parser to handle positional arguments correctly
also made the "2 spaces" prefix for some labels explicit to the use-case
(this should actually get moved elsewhere)
Peter Stamfest [Fri, 5 Sep 2014 06:39:42 +0000 (08:39 +0200)]
Fix tune related test scripts to work in the presence of a caching daemon.
Also do explicit "forgetting" of just tuned RRD, because "tune" is
never done via rrdcached.
Peter Stamfest [Thu, 4 Sep 2014 06:50:42 +0000 (08:50 +0200)]
fetch data from cached with full double precision to make sure results
obtained with and without rrdcached are the same.
This might be deemed to high a cost, because the line protocol now carries
up to 70% more data
Peter Stamfest [Thu, 4 Sep 2014 06:46:29 +0000 (08:46 +0200)]
Fix rrd create via cached - file name sanitation (absolutation?)
did quietly assume that files would have to exists. Which ist not
quite right for many "create" usecases.
Peter Stamfest [Tue, 2 Sep 2014 15:07:53 +0000 (17:07 +0200)]
implement new test "style": valgrind-logfile. This dumps everything to
log files named <test>-valgrind.log and only reports the LEAK SUMMARY to
stderr in case of reported errors. On travis, currently the logfiles
are effectively lost, but this all avoids problems with total test outputs
exceeding the current travis limit of 4MB...
Peter Stamfest [Wed, 27 Aug 2014 06:28:01 +0000 (08:28 +0200)]
add custom quicksort, where the comparison function takes an extra argument for parameters
Using quicksort might in practice actually be overkill, because most of the time we will
sort rather short arrays anyway.
Peter Stamfest [Thu, 21 Aug 2014 10:12:05 +0000 (12:12 +0200)]
first real prefill implementation
- this still contains loads of debug output and some real problems and
assumptions, so it won't work correctly for most cases
Peter Stamfest [Wed, 20 Aug 2014 22:43:18 +0000 (00:43 +0200)]
Move custom logic row-rotation logic from write_fh, originally there to
support rrd_restore with its random cur_row placement to
rrd_restore itself. This makes write_fh a truly reuseable function.
Peter Stamfest [Thu, 14 Aug 2014 07:05:04 +0000 (09:05 +0200)]
Keep a pointer to the internal rrd data structure with rrd_file_t objects
to simplify management of opened RRD files - now it is only required
to keep the rrd_file_t object around to be able to manage the underlying
rrd structure and not both objects in some additional data structure.
Peter Stamfest [Wed, 13 Aug 2014 18:14:31 +0000 (20:14 +0200)]
fix init_cdp. The original code implicitly always used the first pdp_prep
structure of the underlying rrd, which lead to wrong cdp initialization in
experimental code. This change will be needed for future work.
Peter Stamfest [Wed, 13 Aug 2014 18:10:43 +0000 (20:10 +0200)]
move write_rrd and write_fh into rrd_create.c.
While this is purely cosmetic, writing an RRD file is actually creating one, so
it makes sense to concentrate the various writing functions into one place.
petgit [Wed, 27 Aug 2014 19:52:53 +0000 (21:52 +0200)]
use of g_free() cause heap damage on windows if memory is allocated by malloc(). Key is allocated by malloc() in sprintf_alloc(), so free() should use.
Peter A. Bigot [Tue, 5 Aug 2014 21:39:58 +0000 (16:39 -0500)]
configure.ac: remove explicit default path for libxml2
OpenEmbedded needs to cross-compile rrdtool, and historically has
applied this patch locally because the implementation of EX_CHECK_ALL
unconditionally added -I/usr/include. Since the correct include path is
automatically provided by a successful pkg-config invocation there is no
need to set it explicitly.