Age | Commit message (Collapse) | Author |
|
|
|
|
|
|
|
|
|
* The cmdline behavior has been slightly tweaked to facilitate changing
behavior based on the current working directory.
- If NOT in a subdirectory of packages and no package name is supplied
and the action is not clean or distclean, then assume the target is
<project>/all.
- If NOT in a a subdirectory of packages and the action is clean or
distclean, then assume we are cleaning/distcleaning everything
regardless.
- If in a category directory and no target is specified, assume
a target of <category>/all for all actions.
- If in a package and no target is specified, assume
<category>/<package> for all actions.
|
|
* The source command is useful for developers allowing them to checkout package
sources into 'packages/<CATEGORY>/<NAME>/source. This is particularly useful
for developing on sources maintained within an SCM.
|
|
|
|
* Really the current README and scripts need to be pushed into .builder/ so
they stop polluting the core of the tree.
* A lot of the README is more about design philosiphy of the build-engine and
not suitable for a README file.
|
|
|
|
|
|
* Finally support defining pkg_prep() in the Buildrules and calling it from
build-package. This has the potential to cleanup some convoluted build
scripts.
|
|
* Do not simply overwrite the package as that can leave old artifacts laying
around the sysroot. Instead we now destroy the old package via the package
catalog before installing the binary archive.
|
|
* Itterate the directories using a for loop
* Do not create an 'S' directory as most archives extract to that directory
name.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
* Allow packages to clean themselves before purging sysroot/
|
|
* This allows the idea of "empty" virtual packages and glue packages which
combine the artifacts of other packages.
|
|
|
|
|
|
* Attempting to call build-clean from build-distclean has some undesirable
results. It may be possible to do this in the future by supplying a
--distclean option to build-clean, but for now that isn't possible.
|
|
|
|
|
|
|
|
* We already called the package cleanup which already cleaned out the workdir,
there was no reason to clobber the tmpdir as well. More importantly, there
is the very real possibility that our TMPDIR is not within the build-tree, it
may be in /tmp, and clobbering /tmp/* would be bad.
|
|
|
|
* Only report that the package is being cleaned IF it is actually being
cleaned.
|
|
* We should only be recording files and removing files. Purging
empty-directories can be done as a second pass.
|
|
|
|
|
|
|
|
|
|
* While trapping ERR works on bash, it does not work within ash.
Made some minor fixups to remove trapping ERR
* Bugfix .. source the environment in before attempting to use package
specific variables.
|
|
|
|
* While most systems will support CHOST prefixed cc/c++ commands, many
of them do not support prefixing commands like 'ar', so test for the
commands before exporting them.
|
|
|
|
* Reworked the core configuration system so that per category/package
rules can inherit arbitrary data from the project config (necessary
for implementing toolchain rules).
* Copied the GNU config.guess script into the system as
build-dumpmachine to facilitate detecting the CBUILD.
* Removed arbitrary variables previously used for CHOST construction.
It is simpler to set the CHOST and ARCH as opposed to setting ARCH,
VENDOR, LIBC, PLATFORM, etc...
* Cleaned up some variables names
|
|
* The binpkg index file is a 0 delimited file. Mut call xargs with -0
for it to parse the stream correctly.
|
|
* Simply bring the variable inline with variable naming convention.
|
|
* A packages binpkg file has a timestamp that is older than the
containing tarball. After extracging a packages binpkg tarball we
need to touch the /var/db/binpkg marker so that Make is happy.
|
|
|
|
* BUG we should be setting CBUILD if it is unset, though I suppose that
may come at a later time still.
|
|
|
|
|
|
* Not all platforms attach a libc API to the CHOST name. If a LIBC API
was specified then use it, else ignore it.
|
|
* Have cleaned up a number of the hardcoded behavior rules surrounding
the toolchain. Instead of having specialized toolchain rules, a
package or category of packages can simply be flagged as NOCLEAN, such
that the 'build clean' command will skip over them.
* Migrate the 'installed' marker out of the temp path and into the
SYSROOT. This allows 'build clean' to clobber TMP w/out effecting
stateful data.
* Reintroduced the distclean command which ignores NOCLEAN.
|
|
|