| Message ID | 20231208101522.198832-4-alex@linutronix.de |
|---|---|
| State | New |
| Headers | show |
| Series | [1/6] selftest/sstatetest: print output from bitbake with actual newlines, not \n | expand |
On Fri, 2023-12-08 at 11:15 +0100, Alexander Kanavin wrote: > noexec flag has an unfortunate side effect of not writing out the .siginfo > file into sstate (because that is done in sstate bbclass > from a task-completed event handler). > > In the absence of the siginfo file, diffsigs code is unable to trace back > the change in task signatures if the change is really basic and > affects tasks coming ahead of gnu-config-native:do_compile. > > Signed-off-by: Alexander Kanavin <alex@linutronix.de> > --- > meta/recipes-devtools/gnu-config/gnu-config_git.bb | 3 ++- > 1 file changed, 2 insertions(+), 1 deletion(-) > > diff --git a/meta/recipes-devtools/gnu-config/gnu-config_git.bb b/meta/recipes-devtools/gnu-config/gnu-config_git.bb > index 718f798a00e..c72de159ce2 100644 > --- a/meta/recipes-devtools/gnu-config/gnu-config_git.bb > +++ b/meta/recipes-devtools/gnu-config/gnu-config_git.bb > @@ -19,7 +19,8 @@ UPSTREAM_CHECK_COMMITS = "1" > > CLEANBROKEN = "1" > > -do_compile[noexec] = "1" > +do_compile () { > +} > > do_install () { > install -d ${D}${datadir}/gnu-config \ > You have a good point here but by this logic, we need to remove *all* noexec tasks. We don't really want to do that and I'm not sure we want to set this precedent. I wonder if we could just remove the compile task entirely and what that would break... Cheers, Richard
> -----Original Message----- > From: openembedded-core@lists.openembedded.org <openembedded-core@lists.openembedded.org> On Behalf Of Richard Purdie > Sent: den 8 december 2023 12:35 > To: Alexander Kanavin <alex.kanavin@gmail.com>; openembedded-core@lists.openembedded.org > Cc: Alexander Kanavin <alex@linutronix.de> > Subject: Re: [OE-core] [PATCH 4/6] gnu-config: add a do_compile task > > On Fri, 2023-12-08 at 11:15 +0100, Alexander Kanavin wrote: > > noexec flag has an unfortunate side effect of not writing out the .siginfo > > file into sstate (because that is done in sstate bbclass > > from a task-completed event handler). > > > > In the absence of the siginfo file, diffsigs code is unable to trace back > > the change in task signatures if the change is really basic and > > affects tasks coming ahead of gnu-config-native:do_compile. > > > > Signed-off-by: Alexander Kanavin <alex@linutronix.de> > > --- > > meta/recipes-devtools/gnu-config/gnu-config_git.bb | 3 ++- > > 1 file changed, 2 insertions(+), 1 deletion(-) > > > > diff --git a/meta/recipes-devtools/gnu-config/gnu-config_git.bb b/meta/recipes-devtools/gnu-config/gnu-config_git.bb > > index 718f798a00e..c72de159ce2 100644 > > --- a/meta/recipes-devtools/gnu-config/gnu-config_git.bb > > +++ b/meta/recipes-devtools/gnu-config/gnu-config_git.bb > > @@ -19,7 +19,8 @@ UPSTREAM_CHECK_COMMITS = "1" > > > > CLEANBROKEN = "1" > > > > -do_compile[noexec] = "1" > > +do_compile () { > > +} > > > > do_install () { > > install -d ${D}${datadir}/gnu-config \ > > > > You have a good point here but by this logic, we need to remove *all* > noexec tasks. We don't really want to do that and I'm not sure we want > to set this precedent. > > I wonder if we could just remove the compile task entirely and what > that would break... > > Cheers, > > Richard Would it be possible to change the code so that the .siginfo files are still produced for noexec tasks? Otherwise it seems using noexec tasks is a real blocker for being able to trace the signatures for any subsequent task. //Peter
On Fri, 2023-12-08 at 23:46 +0000, Peter Kjellerstedt wrote: > > -----Original Message----- > > From: openembedded-core@lists.openembedded.org > > <openembedded-core@lists.openembedded.org> On Behalf Of Richard > > Purdie > > Sent: den 8 december 2023 12:35 > > To: Alexander Kanavin <alex.kanavin@gmail.com>; > > openembedded-core@lists.openembedded.org > > Cc: Alexander Kanavin <alex@linutronix.de> > > Subject: Re: [OE-core] [PATCH 4/6] gnu-config: add a do_compile > > task > > > > On Fri, 2023-12-08 at 11:15 +0100, Alexander Kanavin wrote: > > > noexec flag has an unfortunate side effect of not writing out the > > > .siginfo > > > file into sstate (because that is done in sstate bbclass > > > from a task-completed event handler). > > > > > > In the absence of the siginfo file, diffsigs code is unable to > > > trace back > > > the change in task signatures if the change is really basic and > > > affects tasks coming ahead of gnu-config-native:do_compile. > > > > > > Signed-off-by: Alexander Kanavin <alex@linutronix.de> > > > --- > > > meta/recipes-devtools/gnu-config/gnu-config_git.bb | 3 ++- > > > 1 file changed, 2 insertions(+), 1 deletion(-) > > > > > > diff --git a/meta/recipes-devtools/gnu-config/gnu-config_git.bb > > > b/meta/recipes-devtools/gnu-config/gnu-config_git.bb > > > index 718f798a00e..c72de159ce2 100644 > > > --- a/meta/recipes-devtools/gnu-config/gnu-config_git.bb > > > +++ b/meta/recipes-devtools/gnu-config/gnu-config_git.bb > > > @@ -19,7 +19,8 @@ UPSTREAM_CHECK_COMMITS = "1" > > > > > > CLEANBROKEN = "1" > > > > > > -do_compile[noexec] = "1" > > > +do_compile () { > > > +} > > > > > > do_install () { > > > install -d ${D}${datadir}/gnu-config \ > > > > > > > You have a good point here but by this logic, we need to remove > > *all* > > noexec tasks. We don't really want to do that and I'm not sure we > > want > > to set this precedent. > > > > I wonder if we could just remove the compile task entirely and what > > that would break... > > > > Cheers, > > > > Richard > > Would it be possible to change the code so that the .siginfo files > are > still produced for noexec tasks? Otherwise it seems using noexec > tasks > is a real blocker for being able to trace the signatures for any > subsequent task. Producing the siginfo files requires a full data store and bitbake doesn't retain that information for every recipe, just a summary of the data it needs. It is therefore currently produced in the task execution context currently. With a noexec task, the win is that we never need that context. So this does need some thought. Cheers, Richard
diff --git a/meta/recipes-devtools/gnu-config/gnu-config_git.bb b/meta/recipes-devtools/gnu-config/gnu-config_git.bb index 718f798a00e..c72de159ce2 100644 --- a/meta/recipes-devtools/gnu-config/gnu-config_git.bb +++ b/meta/recipes-devtools/gnu-config/gnu-config_git.bb @@ -19,7 +19,8 @@ UPSTREAM_CHECK_COMMITS = "1" CLEANBROKEN = "1" -do_compile[noexec] = "1" +do_compile () { +} do_install () { install -d ${D}${datadir}/gnu-config \
noexec flag has an unfortunate side effect of not writing out the .siginfo file into sstate (because that is done in sstate bbclass from a task-completed event handler). In the absence of the siginfo file, diffsigs code is unable to trace back the change in task signatures if the change is really basic and affects tasks coming ahead of gnu-config-native:do_compile. Signed-off-by: Alexander Kanavin <alex@linutronix.de> --- meta/recipes-devtools/gnu-config/gnu-config_git.bb | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)