[OE-core] [PATCH] gconf.bbclass: make postinstall fail silently when running offline

Burton, Ross ross.burton at intel.com
Fri Feb 22 16:35:52 UTC 2013


Hi,

On 22 February 2013 14:58, Laurentiu Palcu <laurentiu.palcu at intel.com> wrote:
> I didn't write GConf. I don't know the reason why they chose to declare
> the following list of characters invalid:
> "\t\r\n\"$&<>,+=#!()'|{}[]?~`;%\\".
>
> I did look in the commit message and found no other explanations except
> this:
> "gconf/gconf-backend.c: (gconf_address_valid), (gconf_get_backend):
> check the backend address doesn't contain any special characters."
>
> Here is the entire commit:
> http://git.gnome.org/browse/gconf/commit/?id=3d720f4a0c00af31c1d53fc4aa45d6d6580c433e
>
> OK, let's say I remove '+' from that list. I agree, it doesn't look like
> it would have bad consequences but what happens if, in the future, the
> path contains another "invalid" character? Where do we draw the line?

+ is notable in that list because it's not handled specially by the
shell, whereas practictically the rest of them are.

> There was a patchset on the ML these days that enabled postinstall
> output redirection to a certain file. Why not have this activated all
> the time for the native dpkg/opkg/rpm and, on request, for the target
> binaries? We could inspect the logs in case some postinstalls failed to
> run on host but we don't end the build since, maybe, some people would
> still be fine with running those postinstalls on target.

Having those logs is useful for inspection, but as we're trying to run
all postinsts on the host this trivially-triggered problem needs to be
resolved.  Patching the sanity test is a reasonable solution.

Ross




More information about the Openembedded-core mailing list