At least on baldras' version of screen, the stuff command argument here
absolutely needs to be wrapped in single or double quotes to work.
I actually fixed this months ago but forgot to commit.
* Using scons -Y to point scons to a source tree that might be
read-only for the user building campaignd/wesnothd (as is the case on
baldras).
* Renamed or changed several variables for clarity.
* As a side-effect, links within the server instance dir are now
absolute rather than relative (/home/wesnoth/build/foo/bar instead of
../../build/foo/bar).
* ~/builds-int/<version> is the intermediate build directory
* ~/builds/<build-id> remains as the target directory
* /usr/src/wesnoth is the source directory base (we need to handle
basilic and gonzo separately later)
I'm fairly sure nobody other than me could possibly demand a more
detailed description of this commit.
They mostly seem to be code cleanups or style changes, plus the hack to
try to start the irc echo for the old instance after restarting... which
doesn't seem to be working.
* Added code for handling 1.8 and 1.10
* Don't use libana for Wesnoth versions that include it (1.9.x)
* We now use Git, not SVN. The downstream repository is managed
externally, so we need to update a git-new-workdir checkout every
time we want to build.
* Other nondescript changes
(Apologies for the horrible commit message, but it looks like there was
more than my own changes awaiting to be committed.)
I wrote this a long time ago and have been using it to ease testing of
Wesnoth's default configuration under different conditions ever since.
It creates a new configuration and data directory under /tmp that only
exists for the duration of the Wesnoth session, and disables WML cache
functionality.
This is extremely useful for quickly testing and debugging functionality
dependant on user configuration settings.
Overall statistics (only for files with a smaller recompressed size):
Original size: 16720 KiB on 404 files
Optimized size: 15237 KiB
Total saving: 1482 KiB = 8% decrease
Most references to and dependencies on Subversion have been removed.
"+svn" is now "+dev". Files that can't be fixed yet have a FIXME-GIT
comment in them; most of these are in the website tests.
...when there are previous combats to account for.
(It passes the attack simulator check.)
The basic problem with the old system is that it treated the number of
strikes as an independent variable, even though it depends on the
number of hit points.
There is some additional overhead with this method, but I optimized
some things for the common case (no swarm), and benchmarking indicated
a negligible 1% increase in run time when no units had swarm.
...and benchmarking.
Biggest change: the first unit, which has all checked specials, gets
to actually attack.
(Previously, its swarm special and low hit points resulted in no attacks.)
...and verification.
The old set had several deficiencies, such as every unit with slow
also having only one attack and having the swarm special (so usually
no attacks). This set keeps the specials and stats on separate cycles.
