| Message ID | 20240223121040.1608421-1-m.felsch@pengutronix.de |
|---|---|
| State | New |
| Headers | show |
| Series | devicetree.bbclass: fix do_fetch stage | expand |
On Fri, 2024-02-23 at 13:10 +0100, Marco Felsch wrote: > The fetch stage should always start with a clean ${S} directory to > not > try to compile leftovers from previous builds. > > Signed-off-by: Marco Felsch <m.felsch@pengutronix.de> > --- > meta/classes-recipe/devicetree.bbclass | 2 ++ > 1 file changed, 2 insertions(+) > > diff --git a/meta/classes-recipe/devicetree.bbclass b/meta/classes- > recipe/devicetree.bbclass > index bd50d7fa1d38..655460a9dbef 100644 > --- a/meta/classes-recipe/devicetree.bbclass > +++ b/meta/classes-recipe/devicetree.bbclass > @@ -77,6 +77,8 @@ python () { > d.appendVarFlag("do_compile", "depends", " > virtual/kernel:do_configure") > } > > +do_fetch[cleandirs] = "${S}" > + > def expand_includes(varname, d): > import glob > includes = set() > This doesn't make sense. do_fetch works with the fetcher in directories outside WORKDIR. The sources only become available at do_unpack. The correct task would therefore be unpack. This is then covered by base.bbclass: meta/classes-global/base.bbclass:do_unpack[cleandirs] = "${@d.getVar('S') if os.path.normpath(d.getVar('S')) != os.path.normpath(d.getVar('WORKDIR')) else os.path.join('${S}', 'patches')}" Cheers, Richard
On 24-02-23, Richard Purdie wrote: > On Fri, 2024-02-23 at 13:10 +0100, Marco Felsch wrote: > > The fetch stage should always start with a clean ${S} directory to > > not > > try to compile leftovers from previous builds. > > > > Signed-off-by: Marco Felsch <m.felsch@pengutronix.de> > > --- > > �meta/classes-recipe/devicetree.bbclass | 2 ++ > > �1 file changed, 2 insertions(+) > > > > diff --git a/meta/classes-recipe/devicetree.bbclass b/meta/classes- > > recipe/devicetree.bbclass > > index bd50d7fa1d38..655460a9dbef 100644 > > --- a/meta/classes-recipe/devicetree.bbclass > > +++ b/meta/classes-recipe/devicetree.bbclass > > @@ -77,6 +77,8 @@ python () { > > �������� d.appendVarFlag("do_compile", "depends", " > > virtual/kernel:do_configure") > > �} > > � > > +do_fetch[cleandirs] = "${S}" > > + > > �def expand_includes(varname, d): > > ���� import glob > > ���� includes = set() > > > > This doesn't make sense. do_fetch works with the fetcher in directories > outside WORKDIR. The sources only become available at do_unpack. The > correct task would therefore be unpack. Good to know, what I inspect is that old dts(i) files are still present once I adapted the SRC_URI path e.g: SRC_URI = "a.dts b.dts" to SRC_URI = "b.dts" and so this class tries to compile both, albeit I explicite removed a.dts from SRC_URI. Regards, Marco > > This is then covered by base.bbclass: > > meta/classes-global/base.bbclass:do_unpack[cleandirs] = "${@d.getVar('S') if os.path.normpath(d.getVar('S')) != os.path.normpath(d.getVar('WORKDIR')) else os.path.join('${S}', 'patches')}" > > Cheers, > > Richard > >
On Fri, 2024-02-23 at 14:01 +0100, Marco Felsch wrote: > On 24-02-23, Richard Purdie wrote: > > On Fri, 2024-02-23 at 13:10 +0100, Marco Felsch wrote: > > > The fetch stage should always start with a clean ${S} directory > > > to > > > not > > > try to compile leftovers from previous builds. > > > > > > Signed-off-by: Marco Felsch <m.felsch@pengutronix.de> > > > --- > > > meta/classes-recipe/devicetree.bbclass | 2 ++ > > > 1 file changed, 2 insertions(+) > > > > > > diff --git a/meta/classes-recipe/devicetree.bbclass > > > b/meta/classes- > > > recipe/devicetree.bbclass > > > index bd50d7fa1d38..655460a9dbef 100644 > > > --- a/meta/classes-recipe/devicetree.bbclass > > > +++ b/meta/classes-recipe/devicetree.bbclass > > > @@ -77,6 +77,8 @@ python () { > > > d.appendVarFlag("do_compile", "depends", " > > > virtual/kernel:do_configure") > > > } > > > > > > +do_fetch[cleandirs] = "${S}" > > > + > > > def expand_includes(varname, d): > > > import glob > > > includes = set() > > > > > > > This doesn't make sense. do_fetch works with the fetcher in > > directories > > outside WORKDIR. The sources only become available at do_unpack. > > The > > correct task would therefore be unpack. > > Good to know, what I inspect is that old dts(i) files are still > present > once I adapted the SRC_URI path e.g: > > SRC_URI = "a.dts b.dts" > > to > > SRC_URI = "b.dts" > > and so this class tries to compile both, albeit I explicite removed > a.dts from SRC_URI. This is a known problem, I thought there was an open bug for it but I can't seem to find it right now. Basically the problem is that the fetcher can't know what it previously placed during do_unpack and know what it needs to clean up. Deleting the contents of ${S} sounds straight forward but in reality, it isn't quite that simple, particularly due to S = ${WORKDIR} that some recipes use. Longer term the plan is to drop support for S = ${WORKDIR} and then we can simplify things but nobody has got around to doing that work yet. Cheers, Richard
On 24-02-23, Marco Felsch wrote: > On 24-02-23, Richard Purdie wrote: > > On Fri, 2024-02-23 at 13:10 +0100, Marco Felsch wrote: > > > The fetch stage should always start with a clean ${S} directory to > > > not > > > try to compile leftovers from previous builds. > > > > > > Signed-off-by: Marco Felsch <m.felsch@pengutronix.de> > > > --- > > > �meta/classes-recipe/devicetree.bbclass | 2 ++ > > > �1 file changed, 2 insertions(+) > > > > > > diff --git a/meta/classes-recipe/devicetree.bbclass b/meta/classes- > > > recipe/devicetree.bbclass > > > index bd50d7fa1d38..655460a9dbef 100644 > > > --- a/meta/classes-recipe/devicetree.bbclass > > > +++ b/meta/classes-recipe/devicetree.bbclass > > > @@ -77,6 +77,8 @@ python () { > > > �������� d.appendVarFlag("do_compile", "depends", " > > > virtual/kernel:do_configure") > > > �} > > > � > > > +do_fetch[cleandirs] = "${S}" > > > + > > > �def expand_includes(varname, d): > > > ���� import glob > > > ���� includes = set() > > > > > > > This doesn't make sense. do_fetch works with the fetcher in directories > > outside WORKDIR. The sources only become available at do_unpack. The > > correct task would therefore be unpack. > > Good to know, what I inspect is that old dts(i) files are still present > once I adapted the SRC_URI path e.g: > > SRC_URI = "a.dts b.dts" > > to > > SRC_URI = "b.dts" > > and so this class tries to compile both, albeit I explicite removed > a.dts from SRC_URI. > > Regards, > Marco > > > > > > > This is then covered by base.bbclass: > > > > meta/classes-global/base.bbclass:do_unpack[cleandirs] = "${@d.getVar('S') if os.path.normpath(d.getVar('S')) != os.path.normpath(d.getVar('WORKDIR')) else os.path.join('${S}', 'patches')}" To answer myself, the issue is that the class does the following: S = "${WORKDIR}" so the cleandirs command point to the wrong dir. Regards, Marco
On 24-02-23, Richard Purdie wrote: > On Fri, 2024-02-23 at 14:01 +0100, Marco Felsch wrote: > > On 24-02-23, Richard Purdie wrote: > > > On Fri, 2024-02-23 at 13:10 +0100, Marco Felsch wrote: > > > > The fetch stage should always start with a clean ${S} directory > > > > to > > > > not > > > > try to compile leftovers from previous builds. > > > > > > > > Signed-off-by: Marco Felsch <m.felsch@pengutronix.de> > > > > --- > > > > �meta/classes-recipe/devicetree.bbclass | 2 ++ > > > > �1 file changed, 2 insertions(+) > > > > > > > > diff --git a/meta/classes-recipe/devicetree.bbclass > > > > b/meta/classes- > > > > recipe/devicetree.bbclass > > > > index bd50d7fa1d38..655460a9dbef 100644 > > > > --- a/meta/classes-recipe/devicetree.bbclass > > > > +++ b/meta/classes-recipe/devicetree.bbclass > > > > @@ -77,6 +77,8 @@ python () { > > > > �������� d.appendVarFlag("do_compile", "depends", " > > > > virtual/kernel:do_configure") > > > > �} > > > > � > > > > +do_fetch[cleandirs] = "${S}" > > > > + > > > > �def expand_includes(varname, d): > > > > ���� import glob > > > > ���� includes = set() > > > > > > > > > > This doesn't make sense. do_fetch works with the fetcher in > > > directories > > > outside WORKDIR. The sources only become available at do_unpack. > > > The > > > correct task would therefore be unpack. > > > > Good to know, what I inspect is that old dts(i) files are still > > present > > once I adapted the SRC_URI path e.g: > > > > SRC_URI = "a.dts b.dts" > > > > to > > > > SRC_URI = "b.dts" > > > > and so this class tries to compile both, albeit I explicite removed > > a.dts from SRC_URI. > > This is a known problem, I thought there was an open bug for it but I > can't seem to find it right now. > > Basically the problem is that the fetcher can't know what it previously > placed during do_unpack and know what it needs to clean up. Deleting > the contents of ${S} sounds straight forward but in reality, it isn't > quite that simple, particularly due to S = ${WORKDIR} that some recipes > use. > > Longer term the plan is to drop support for S = ${WORKDIR} and then we > can simplify things but nobody has got around to doing that work yet. Okay, thanks for input. In that case I will wait and notice it as known bug. Regards, Marco
diff --git a/meta/classes-recipe/devicetree.bbclass b/meta/classes-recipe/devicetree.bbclass index bd50d7fa1d38..655460a9dbef 100644 --- a/meta/classes-recipe/devicetree.bbclass +++ b/meta/classes-recipe/devicetree.bbclass @@ -77,6 +77,8 @@ python () { d.appendVarFlag("do_compile", "depends", " virtual/kernel:do_configure") } +do_fetch[cleandirs] = "${S}" + def expand_includes(varname, d): import glob includes = set()
The fetch stage should always start with a clean ${S} directory to not try to compile leftovers from previous builds. Signed-off-by: Marco Felsch <m.felsch@pengutronix.de> --- meta/classes-recipe/devicetree.bbclass | 2 ++ 1 file changed, 2 insertions(+)