it contains a list with all triggers for all subpackages.

separate trigger script from path list with '='
separate paths in path list with ':'

example:

pkgname="foo"
subpackages="foo-libs foo-dev"
triggers="foo.trigger=/usr:/usr/lib/foo
	foo-libs.trigger=/usr:/usr/share/foo"

We need to list all install scripts in $install rather than specify it
in subpackage

we check early that suffix is valid and that we have a matching
pkgname or subpackage

We don't really need to see what packages gets uninstalled, but error
messages might be nice to see. So we are quiet when uninstalling.

only main package has the trigger

we need set the $CARCH in the package destination
needs also to be passed over for recursive action

We no require apk-tools-2.1

apk-tools-2.0's apk index will not show the arch so we need to use
tar to find arch if we have old apk. We prefer use apk index since
it will only read first block of file, regardless size, while tar
will read entire file. Reading entire file is slow if file is big.

So we check apk --version and use tar if needed, and apk index if
possible.

This is so apk-tools-2.1 works

we need use "$srcdir"

So SUDO="sudo -E" works for http proxy settings

Jeremy Thomerson authored 14 years ago and Natanael Copa committed 14 years ago

Currently, if the build server downloads a file that fails the md5sum check
for some reason (perhaps interrupted download, etc), the file is never deleted
so the build can not succeed until someone manually deletes the bad file.  This
change deletes any file that is a remote source file / package that fails the
md5sum check.  The subsequent rebuild will then be forced to download the file
again.

Our build infrastruc needs to handle it

We can either set maintainer="..." in APKBUILD or use it as comment.

This can be useful for package browser.

due to unintialized and non-local variable the tracing of deps
didnt work as expected when dependencies was provided by the apkbuild
itself. (libgcc ended up depending on libgcj which depended on binutils)

and we dont want -dev packages to be noarch

this check verifies that the *.initd scripts are #!/sbin/runscript

This subcommand verifies if upstream sources still exists and echoes
an error if it does not.

This is supposed to be used from a script that checks the validity
of source regularily.

Matt Smith authored 14 years ago and Natanael Copa committed 14 years ago

Removed the asterisk from the beginning of the https check, and moved the
saveas- check above.

Matt Smith authored 14 years ago and Natanael Copa committed 14 years ago

'saveas-*://' URI support has been created for use with the source= line of
APKBUILD files.

It allows for a remote source file to be saved with an arbitrary filename.  This
is useful in situations where the last component of the URI is not the preferred
filename.

Here's how it works.  Say we have the following URI:

  http://oss.example.org/?get=software&ver=1.0

Both Busybox Wget and GNU Wget will save this with the filename:

  ?get=software&ver=1.0

To get around this, we could use cURL to save the file using the filename in the
HTTP response headers:

  $ curl -JO "http://oss.example.org/?get=software&ver=1.0"

Or we could use this 'saveas' hack.  Essentially, the original URI is converted
to read:

  saveas-http://oss.example.org/?get=software&ver=1.0/software-1.0.tar.gz

In the download process, the 'saveas-' portion is removed, and the file is
downloaded from the original URI, but is saved with the filename being the last
component of the URI.  In this case, it will be saved as 'software-1.0.tar.gz'.

It is designed so that it works with any protocol supported by abuild.
For example:

  saveas-ftp://oss.example.org/?get=software&ver=1.0/software-1.0.tar.gz

Check it out and let me know what you think.

Thanks,
Matt

we now fail if noarch is set wrong