| Message ID | 20220701020939.62295-1-alhe@linux.microsoft.com |
|---|---|
| State | Accepted, archived |
| Commit | a887bd96fd0a15398e8077ea79df5070971866e4 |
| Headers | show |
| Series | package.bbclass: Fix base directory for debugsource files when using externalsrc | expand |
Hi Alejandro, Alejandro Hernandez Samaniego <alhe@linux.microsoft.com> escreveu no dia sexta, 1/07/2022 à(s) 03:09: > While executing do_package, bitbake checks for a list of > debug source files and uses a pattern to match the ones > to be included in copydebugsources. > > Previously when externalsrc was in use either directly or by > using devtool, the source location changed and this pattern > no longer matched, hence debug source files failed to be > included in the corresponding package. > > Check when the source directory isnt the default (based on > WORKDIR), and change the pattern used to match debug source > files if that is the case, allowing us to perform do_package > properly. > > Workaround debugsource.list containing paths from the host by > moving debug source files away from the host directory > structure to avoid host contamination (this seems to happen > when packages use $TMPDIR/work-shared and externalsrc is > in use). > > Test matrix included using: > - devtool to use externalsrc automatically > - externalsrc with a non-devtool based source directory > - No externalsrc at all > Tested the following packages to be working: > - glibc ($TMPDIR/work-shared based) > - libxcrypt ($TMPDIR/work based) > > [YOCTO 8015] > > Signed-off-by: Alejandro Enedino Hernandez Samaniego < > alhe@linux.microsoft.com> > --- > meta/classes/package.bbclass | 31 ++++++++++++++++++++++++++++--- > 1 file changed, 28 insertions(+), 3 deletions(-) > > diff --git a/meta/classes/package.bbclass b/meta/classes/package.bbclass > index 62050a18b8..4850134022 100644 > --- a/meta/classes/package.bbclass > +++ b/meta/classes/package.bbclass > @@ -553,13 +553,25 @@ def copydebugsources(debugsrcdir, sources, d): > strip = d.getVar("STRIP") > objcopy = d.getVar("OBJCOPY") > workdir = d.getVar("WORKDIR") > + sdir = d.getVar("S") > + sparentdir = os.path.dirname(os.path.dirname(sdir)) > + sbasedir = os.path.basename(os.path.dirname(sdir)) + "/" + > os.path.basename(sdir) > workparentdir = os.path.dirname(os.path.dirname(workdir)) > workbasedir = os.path.basename(os.path.dirname(workdir)) + "/" + > os.path.basename(workdir) > > + # If S isnt based on WORKDIR we can infer our sources are located > elsewhere, > + # e.g. using externalsrc; use S as base for our dirs > This doesn't seem very safe imo, if the idea is to check if the externalsrc in use it is better using: if bb.data.inherits_class('externalsrc', d): Jose + if workdir in sdir: > + basedir = workbasedir > + parentdir = workparentdir > + else: > + basedir = sbasedir > + parentdir = sparentdir > + > # If build path exists in sourcefile, it means toolchain did not > use > # -fdebug-prefix-map to compile > if checkbuildpath(sourcefile, d): > - localsrc_prefix = workparentdir + "/" > + localsrc_prefix = parentdir + "/" > else: > localsrc_prefix = "/usr/src/debug/" > > @@ -581,7 +593,7 @@ def copydebugsources(debugsrcdir, sources, d): > processdebugsrc += "sed 's#%s##g' | " > processdebugsrc += "(cd '%s' ; cpio -pd0mlL --no-preserve-owner > '%s%s' 2>/dev/null)" > > - cmd = processdebugsrc % (sourcefile, workbasedir, > localsrc_prefix, workparentdir, dvar, debugsrcdir) > + cmd = processdebugsrc % (sourcefile, basedir, localsrc_prefix, > parentdir, dvar, debugsrcdir) > try: > subprocess.check_output(cmd, shell=True, > stderr=subprocess.STDOUT) > except subprocess.CalledProcessError: > @@ -591,9 +603,22 @@ def copydebugsources(debugsrcdir, sources, d): > # cpio seems to have a bug with -lL together and symbolic links > are just copied, not dereferenced. > # Work around this by manually finding and copying any symbolic > links that made it through. > cmd = "find %s%s -type l -print0 -delete | sed s#%s%s/##g | (cd > '%s' ; cpio -pd0mL --no-preserve-owner '%s%s')" % \ > - (dvar, debugsrcdir, dvar, debugsrcdir, workparentdir, > dvar, debugsrcdir) > + (dvar, debugsrcdir, dvar, debugsrcdir, parentdir, dvar, > debugsrcdir) > subprocess.check_output(cmd, shell=True, stderr=subprocess.STDOUT) > > + > + # debugsources.list may be polluted from the host if we used > externalsrc, > + # cpio uses copy-pass and may have just created a directory > structure > + # matching the one from the host, if thats the case move those > files to > + # debugsrcdir to avoid host contamination. > + # Empty dir structure will be deleted in the next step. > + > + # Same check as above for externalsrc > + if workdir not in sdir: > + if os.path.exists(dvar + debugsrcdir + sdir): > + cmd = "mv %s%s%s/* %s%s" % (dvar, debugsrcdir, sdir, > dvar,debugsrcdir) > + subprocess.check_output(cmd, shell=True, > stderr=subprocess.STDOUT) > + > # The copy by cpio may have resulted in some empty directories! > Remove these > cmd = "find %s%s -empty -type d -delete" % (dvar, debugsrcdir) > subprocess.check_output(cmd, shell=True, stderr=subprocess.STDOUT) > -- > 2.25.1 > > > -=-=-=-=-=-=-=-=-=-=-=- > Links: You receive all messages sent to this group. > View/Reply Online (#167448): > https://lists.openembedded.org/g/openembedded-core/message/167448 > Mute This Topic: https://lists.openembedded.org/mt/92100558/5052612 > Group Owner: openembedded-core+owner@lists.openembedded.org > Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [ > quaresma.jose@gmail.com] > -=-=-=-=-=-=-=-=-=-=-=- > >
On Fri, 2022-07-01 at 12:05 +0100, Jose Quaresma wrote: > Hi Alejandro, > > Alejandro Hernandez Samaniego <alhe@linux.microsoft.com> escreveu no dia sexta, 1/07/2022 à(s) 03:09: > > While executing do_package, bitbake checks for a list of > > debug source files and uses a pattern to match the ones > > to be included in copydebugsources. > > > > Previously when externalsrc was in use either directly or by > > using devtool, the source location changed and this pattern > > no longer matched, hence debug source files failed to be > > included in the corresponding package. > > > > Check when the source directory isnt the default (based on > > WORKDIR), and change the pattern used to match debug source > > files if that is the case, allowing us to perform do_package > > properly. > > > > Workaround debugsource.list containing paths from the host by > > moving debug source files away from the host directory > > structure to avoid host contamination (this seems to happen > > when packages use $TMPDIR/work-shared and externalsrc is > > in use). > > > > Test matrix included using: > > - devtool to use externalsrc automatically > > - externalsrc with a non-devtool based source directory > > - No externalsrc at all > > Tested the following packages to be working: > > - glibc ($TMPDIR/work-shared based) > > - libxcrypt ($TMPDIR/work based) > > > > [YOCTO 8015] > > > > Signed-off-by: Alejandro Enedino Hernandez Samaniego <alhe@linux.microsoft.com> > > --- > > meta/classes/package.bbclass | 31 ++++++++++++++++++++++++++++--- > > 1 file changed, 28 insertions(+), 3 deletions(-) > > > > diff --git a/meta/classes/package.bbclass b/meta/classes/package.bbclass > > index 62050a18b8..4850134022 100644 > > --- a/meta/classes/package.bbclass > > +++ b/meta/classes/package.bbclass > > @@ -553,13 +553,25 @@ def copydebugsources(debugsrcdir, sources, d): > > strip = d.getVar("STRIP") > > objcopy = d.getVar("OBJCOPY") > > workdir = d.getVar("WORKDIR") > > + sdir = d.getVar("S") > > + sparentdir = os.path.dirname(os.path.dirname(sdir)) > > + sbasedir = os.path.basename(os.path.dirname(sdir)) + "/" + os.path.basename(sdir) > > workparentdir = os.path.dirname(os.path.dirname(workdir)) > > workbasedir = os.path.basename(os.path.dirname(workdir)) + "/" + os.path.basename(workdir) > > > > + # If S isnt based on WORKDIR we can infer our sources are located elsewhere, > > + # e.g. using externalsrc; use S as base for our dirs > > > > > This doesn't seem very safe imo, if the idea is to check if the externalsrc in use it is better using: > > if bb.data.inherits_class('externalsrc', d): The code is generic enough that if the source is outside WORKDIR, it should work so I'm not sure it makes sense to gate this on externalsrc only though... Cheers, Richard
Richard Purdie <richard.purdie@linuxfoundation.org> escreveu no dia sexta, 1/07/2022 à(s) 12:19: > On Fri, 2022-07-01 at 12:05 +0100, Jose Quaresma wrote: > > Hi Alejandro, > > > > Alejandro Hernandez Samaniego <alhe@linux.microsoft.com> escreveu no > dia sexta, 1/07/2022 à(s) 03:09: > > > While executing do_package, bitbake checks for a list of > > > debug source files and uses a pattern to match the ones > > > to be included in copydebugsources. > > > > > > Previously when externalsrc was in use either directly or by > > > using devtool, the source location changed and this pattern > > > no longer matched, hence debug source files failed to be > > > included in the corresponding package. > > > > > > Check when the source directory isnt the default (based on > > > WORKDIR), and change the pattern used to match debug source > > > files if that is the case, allowing us to perform do_package > > > properly. > > > > > > Workaround debugsource.list containing paths from the host by > > > moving debug source files away from the host directory > > > structure to avoid host contamination (this seems to happen > > > when packages use $TMPDIR/work-shared and externalsrc is > > > in use). > > > > > > Test matrix included using: > > > - devtool to use externalsrc automatically > > > - externalsrc with a non-devtool based source directory > > > - No externalsrc at all > > > Tested the following packages to be working: > > > - glibc ($TMPDIR/work-shared based) > > > - libxcrypt ($TMPDIR/work based) > > > > > > [YOCTO 8015] > > > > > > Signed-off-by: Alejandro Enedino Hernandez Samaniego < > alhe@linux.microsoft.com> > > > --- > > > meta/classes/package.bbclass | 31 ++++++++++++++++++++++++++++--- > > > 1 file changed, 28 insertions(+), 3 deletions(-) > > > > > > diff --git a/meta/classes/package.bbclass > b/meta/classes/package.bbclass > > > index 62050a18b8..4850134022 100644 > > > --- a/meta/classes/package.bbclass > > > +++ b/meta/classes/package.bbclass > > > @@ -553,13 +553,25 @@ def copydebugsources(debugsrcdir, sources, d): > > > strip = d.getVar("STRIP") > > > objcopy = d.getVar("OBJCOPY") > > > workdir = d.getVar("WORKDIR") > > > + sdir = d.getVar("S") > > > + sparentdir = os.path.dirname(os.path.dirname(sdir)) > > > + sbasedir = os.path.basename(os.path.dirname(sdir)) + "/" + > os.path.basename(sdir) > > > workparentdir = os.path.dirname(os.path.dirname(workdir)) > > > workbasedir = os.path.basename(os.path.dirname(workdir)) + > "/" + os.path.basename(workdir) > > > > > > + # If S isnt based on WORKDIR we can infer our sources are > located elsewhere, > > > + # e.g. using externalsrc; use S as base for our dirs > > > > > > > > > This doesn't seem very safe imo, if the idea is to check if the > externalsrc in use it is better using: > > > > if bb.data.inherits_class('externalsrc', d): > > The code is generic enough that if the source is outside WORKDIR, it > should work so I'm not sure it makes sense to gate this on externalsrc > only though... > My comment is mainly because I see the externalsrc everywhere in this patch. But as said to make it generic it's better as it is. Jose > Cheers, > > Richard > > >
On Thu, 2022-06-30 at 20:09 -0600, Alejandro Hernandez Samaniego wrote: > While executing do_package, bitbake checks for a list of > debug source files and uses a pattern to match the ones > to be included in copydebugsources. > > Previously when externalsrc was in use either directly or by > using devtool, the source location changed and this pattern > no longer matched, hence debug source files failed to be > included in the corresponding package. > > Check when the source directory isnt the default (based on > WORKDIR), and change the pattern used to match debug source > files if that is the case, allowing us to perform do_package > properly. > > Workaround debugsource.list containing paths from the host by > moving debug source files away from the host directory > structure to avoid host contamination (this seems to happen > when packages use $TMPDIR/work-shared and externalsrc is > in use). > > Test matrix included using: > - devtool to use externalsrc automatically > - externalsrc with a non-devtool based source directory > - No externalsrc at all > Tested the following packages to be working: > - glibc ($TMPDIR/work-shared based) > - libxcrypt ($TMPDIR/work based) > > [YOCTO 8015] > > Signed-off-by: Alejandro Enedino Hernandez Samaniego <alhe@linux.microsoft.com> > --- > meta/classes/package.bbclass | 31 ++++++++++++++++++++++++++++--- > 1 file changed, 28 insertions(+), 3 deletions(-) > > diff --git a/meta/classes/package.bbclass b/meta/classes/package.bbclass > index 62050a18b8..4850134022 100644 > --- a/meta/classes/package.bbclass > +++ b/meta/classes/package.bbclass > @@ -553,13 +553,25 @@ def copydebugsources(debugsrcdir, sources, d): > strip = d.getVar("STRIP") > objcopy = d.getVar("OBJCOPY") > workdir = d.getVar("WORKDIR") > + sdir = d.getVar("S") > + sparentdir = os.path.dirname(os.path.dirname(sdir)) > + sbasedir = os.path.basename(os.path.dirname(sdir)) + "/" + os.path.basename(sdir) > workparentdir = os.path.dirname(os.path.dirname(workdir)) > workbasedir = os.path.basename(os.path.dirname(workdir)) + "/" + os.path.basename(workdir) > > + # If S isnt based on WORKDIR we can infer our sources are located elsewhere, > + # e.g. using externalsrc; use S as base for our dirs > + if workdir in sdir: > + basedir = workbasedir > + parentdir = workparentdir > + else: > + basedir = sbasedir > + parentdir = sparentdir > + > # If build path exists in sourcefile, it means toolchain did not use > # -fdebug-prefix-map to compile > if checkbuildpath(sourcefile, d): > - localsrc_prefix = workparentdir + "/" > + localsrc_prefix = parentdir + "/" > else: > localsrc_prefix = "/usr/src/debug/" > > @@ -581,7 +593,7 @@ def copydebugsources(debugsrcdir, sources, d): > processdebugsrc += "sed 's#%s##g' | " > processdebugsrc += "(cd '%s' ; cpio -pd0mlL --no-preserve-owner '%s%s' 2>/dev/null)" > > - cmd = processdebugsrc % (sourcefile, workbasedir, localsrc_prefix, workparentdir, dvar, debugsrcdir) > + cmd = processdebugsrc % (sourcefile, basedir, localsrc_prefix, parentdir, dvar, debugsrcdir) > try: > subprocess.check_output(cmd, shell=True, stderr=subprocess.STDOUT) > except subprocess.CalledProcessError: > @@ -591,9 +603,22 @@ def copydebugsources(debugsrcdir, sources, d): > # cpio seems to have a bug with -lL together and symbolic links are just copied, not dereferenced. > # Work around this by manually finding and copying any symbolic links that made it through. > cmd = "find %s%s -type l -print0 -delete | sed s#%s%s/##g | (cd '%s' ; cpio -pd0mL --no-preserve-owner '%s%s')" % \ > - (dvar, debugsrcdir, dvar, debugsrcdir, workparentdir, dvar, debugsrcdir) > + (dvar, debugsrcdir, dvar, debugsrcdir, parentdir, dvar, debugsrcdir) > subprocess.check_output(cmd, shell=True, stderr=subprocess.STDOUT) > > + > + # debugsources.list may be polluted from the host if we used externalsrc, > + # cpio uses copy-pass and may have just created a directory structure > + # matching the one from the host, if thats the case move those files to > + # debugsrcdir to avoid host contamination. > + # Empty dir structure will be deleted in the next step. > + > + # Same check as above for externalsrc > + if workdir not in sdir: > + if os.path.exists(dvar + debugsrcdir + sdir): > + cmd = "mv %s%s%s/* %s%s" % (dvar, debugsrcdir, sdir, dvar,debugsrcdir) > + subprocess.check_output(cmd, shell=True, stderr=subprocess.STDOUT) > + > # The copy by cpio may have resulted in some empty directories! Remove these > cmd = "find %s%s -empty -type d -delete" % (dvar, debugsrcdir) > subprocess.check_output(cmd, shell=True, stderr=subprocess.STDOUT) This has merged but unfortunately it introduces some kind of regression. If you build a kernel after this is applied, the kernel source in /usr/src/ for debug information is no longer present. This is why I was seeing build failures for the buildpaths QA test addition whilst Bruce was not since I didn't have this applied in that particular build. We need to find out what broke here and how to fix it. Clearly we need some better QA tests as well to detect this. Cheers, Richard
diff --git a/meta/classes/package.bbclass b/meta/classes/package.bbclass index 62050a18b8..4850134022 100644 --- a/meta/classes/package.bbclass +++ b/meta/classes/package.bbclass @@ -553,13 +553,25 @@ def copydebugsources(debugsrcdir, sources, d): strip = d.getVar("STRIP") objcopy = d.getVar("OBJCOPY") workdir = d.getVar("WORKDIR") + sdir = d.getVar("S") + sparentdir = os.path.dirname(os.path.dirname(sdir)) + sbasedir = os.path.basename(os.path.dirname(sdir)) + "/" + os.path.basename(sdir) workparentdir = os.path.dirname(os.path.dirname(workdir)) workbasedir = os.path.basename(os.path.dirname(workdir)) + "/" + os.path.basename(workdir) + # If S isnt based on WORKDIR we can infer our sources are located elsewhere, + # e.g. using externalsrc; use S as base for our dirs + if workdir in sdir: + basedir = workbasedir + parentdir = workparentdir + else: + basedir = sbasedir + parentdir = sparentdir + # If build path exists in sourcefile, it means toolchain did not use # -fdebug-prefix-map to compile if checkbuildpath(sourcefile, d): - localsrc_prefix = workparentdir + "/" + localsrc_prefix = parentdir + "/" else: localsrc_prefix = "/usr/src/debug/" @@ -581,7 +593,7 @@ def copydebugsources(debugsrcdir, sources, d): processdebugsrc += "sed 's#%s##g' | " processdebugsrc += "(cd '%s' ; cpio -pd0mlL --no-preserve-owner '%s%s' 2>/dev/null)" - cmd = processdebugsrc % (sourcefile, workbasedir, localsrc_prefix, workparentdir, dvar, debugsrcdir) + cmd = processdebugsrc % (sourcefile, basedir, localsrc_prefix, parentdir, dvar, debugsrcdir) try: subprocess.check_output(cmd, shell=True, stderr=subprocess.STDOUT) except subprocess.CalledProcessError: @@ -591,9 +603,22 @@ def copydebugsources(debugsrcdir, sources, d): # cpio seems to have a bug with -lL together and symbolic links are just copied, not dereferenced. # Work around this by manually finding and copying any symbolic links that made it through. cmd = "find %s%s -type l -print0 -delete | sed s#%s%s/##g | (cd '%s' ; cpio -pd0mL --no-preserve-owner '%s%s')" % \ - (dvar, debugsrcdir, dvar, debugsrcdir, workparentdir, dvar, debugsrcdir) + (dvar, debugsrcdir, dvar, debugsrcdir, parentdir, dvar, debugsrcdir) subprocess.check_output(cmd, shell=True, stderr=subprocess.STDOUT) + + # debugsources.list may be polluted from the host if we used externalsrc, + # cpio uses copy-pass and may have just created a directory structure + # matching the one from the host, if thats the case move those files to + # debugsrcdir to avoid host contamination. + # Empty dir structure will be deleted in the next step. + + # Same check as above for externalsrc + if workdir not in sdir: + if os.path.exists(dvar + debugsrcdir + sdir): + cmd = "mv %s%s%s/* %s%s" % (dvar, debugsrcdir, sdir, dvar,debugsrcdir) + subprocess.check_output(cmd, shell=True, stderr=subprocess.STDOUT) + # The copy by cpio may have resulted in some empty directories! Remove these cmd = "find %s%s -empty -type d -delete" % (dvar, debugsrcdir) subprocess.check_output(cmd, shell=True, stderr=subprocess.STDOUT)
While executing do_package, bitbake checks for a list of debug source files and uses a pattern to match the ones to be included in copydebugsources. Previously when externalsrc was in use either directly or by using devtool, the source location changed and this pattern no longer matched, hence debug source files failed to be included in the corresponding package. Check when the source directory isnt the default (based on WORKDIR), and change the pattern used to match debug source files if that is the case, allowing us to perform do_package properly. Workaround debugsource.list containing paths from the host by moving debug source files away from the host directory structure to avoid host contamination (this seems to happen when packages use $TMPDIR/work-shared and externalsrc is in use). Test matrix included using: - devtool to use externalsrc automatically - externalsrc with a non-devtool based source directory - No externalsrc at all Tested the following packages to be working: - glibc ($TMPDIR/work-shared based) - libxcrypt ($TMPDIR/work based) [YOCTO 8015] Signed-off-by: Alejandro Enedino Hernandez Samaniego <alhe@linux.microsoft.com> --- meta/classes/package.bbclass | 31 ++++++++++++++++++++++++++++--- 1 file changed, 28 insertions(+), 3 deletions(-)