Re: [PATCH] image.bbclass: Do not expand variables in IMAGE_CMD vardepsexclude

[Richard Purdie <richard.purdie@linuxfoundation.org>]

On Tue, 2016-09-13 at 00:05 +1000, Jonathan Liu wrote:
> Hi Richard,
> 
> On 12 September 2016 at 20:19, Richard Purdie
> <richard.purdie@linuxfoundation.org> wrote:
> > 
> > On Sun, 2016-09-11 at 22:14 +1000, Jonathan Liu wrote:
> > > 
> > > This fixes tashhash mismatch errors creating images when
> > > IMAGE_CMD
> > > references a variable whose value is always changing even though
> > > the
> > > variable is specified in IMAGE_CMD vardepsexclude.
> > > 
> > > Signed-off-by: Jonathan Liu <net147@gmail.com>
> > > ---
> > >  meta/classes/image.bbclass | 4 ++++
> > >  1 file changed, 4 insertions(+)
> > > 
> > > diff --git a/meta/classes/image.bbclass
> > > b/meta/classes/image.bbclass
> > > index 8c8b8a2..f85ea86 100644
> > > --- a/meta/classes/image.bbclass
> > > +++ b/meta/classes/image.bbclass
> > > @@ -441,6 +441,10 @@ python () {
> > >          localdata.delVar('DATETIME')
> > >          localdata.delVar('TMPDIR')
> > > 
> > > +        vardepsexclude = d.getVarFlag('IMAGE_CMD',
> > > 'vardepsexclude',
> > > True) or ''
> > > +        vardepsexclude += ' ' + (d.getVarFlag('IMAGE_CMD_' +
> > > realt,
> > > 'vardepsexclude', True) or '')
> > > +        for var in vardepsexclude.split():
> > > +            localdata.delVar(var)
> > >          image_cmd = localdata.getVar("IMAGE_CMD", True)
> > >          vardeps.add('IMAGE_CMD_' + realt)
> > >          if image_cmd:
> > I think this is complicating an already fragile situation and it
> > worries me. Can you be more specific about which variable is
> > causing
> > problems here?
> See IMAGEDATESTAMP variable in:
> https://git.yoctoproject.org/cgit/cgit.cgi/meta-raspberrypi/tree/clas
> ses/sdcard_image-rpi.bbclass
> https://github.com/linux-sunxi/meta-sunxi/blob/master/classes/sdcard_
> image-sunxi.bbclass
> 
> You can test by changing to:
> IMAGEDATESTAMP = "${@time.clock()}"

I appreciate this doesn't solve the underlying problem but is there a
reason they can't just use ${DATE} there?

I mention this since DATE/TIME are handled by bitbake specially so that
for example if your build rolls over midnight, it doesn't explode
badly. If your images is generated pre midnight but them processed
afterwards, I can see things working badly for that code...

Cheers,

Richard