[OE-core] [PATCH] smartpm: Don't ignore error if RPM transaction fails without problems

Klauer, Daniel Daniel.Klauer at gin.de
Wed Jun 8 11:35:10 UTC 2016


Hello,

I don't know the details of the attempt-install feature, but it looks like my patch did change the retry behaviour, due to this change in smart/backends/rpm/pm.py (in smart-attempt.patch):
-            if probs and sysconf.has("attempt-install", soft=True):
+            if (probs is not None) and sysconf.has("attempt-install", soft=True):

Unlike before, this if block now sets retry = 1 if len(probs) == 0. It looks to me like the attempt-install feature only wants to retry in case it found file conflicts, and there aren't any other problems - but unfortunately retry is only set back to 0 if it finds something other than file conflicts, the case of len(probs) == 0 is not handled inside the if block.

Is this the problem you're seeing? I think it could be fixed by reverting that one change, maybe like this:
-            if (probs is not None) and sysconf.has("attempt-install", soft=True):
+            # If there are file conflicts, remove conflicting packages from transaction and retry,
+            # unless there are other problems too.
+            if (probs is not None) and (len(probs) != 0) and sysconf.has("attempt-install", soft=True):

Daniel
________________________________________
From: Herve Jourdain <herve.jourdain at neuf.fr>
Sent: Tuesday, June 7, 2016 15:17
To: Klauer, Daniel; openembedded-core at lists.openembedded.org
Subject: RE: [OE-core] [PATCH] smartpm: Don't ignore error if RPM transaction fails without problems

Hi Daniel,

I've just met a problem linked to that patch...
Indeed, it appears that your modifications make the original patch work as
intended: check if there are problems like RPMPROB_NEW_FILE_CONFLICT or
RPMPROB_FILE_CONFLICT, then remove the packages from the list of packages to
apply (because it's attempt-install), then try the same transaction again...

The problem is: if not all the packages have problems, then the packages
that didn't have problems installing in the first place will now get an
error: "package already installed"!!!
And since it's not one of the problems that are handled in pm.py, it will
just fail...

What do you think would be the best way to solve this: add a check for all
the codes linked to package already installed in the "attempt-only" check in
pm.py?
Or try to cancel the whole previous transaction - and then keep on
committing the new one?

I can generate a new patch if you want, after your guidance, or let you fix
it, like you prefer.
I do have an environment where I can reproduce this problem easily enough.

Herve

-----Original Message-----
From: openembedded-core-bounces at lists.openembedded.org
[mailto:openembedded-core-bounces at lists.openembedded.org] On Behalf Of
Klauer, Daniel
Sent: mardi 17 mai 2016 14:58
To: openembedded-core at lists.openembedded.org
Subject: [OE-core] [PATCH] smartpm: Don't ignore error if RPM transaction
fails without problems

SmartPM could misinterpret RPM transaction error as success, if ts.run()
(RPM Python API) returns an empty problems list.

This could happen for example if the RPM database is partially corrupted
such that the transaction does not have any problems like conflicts or
missing dependencies, but still can't be committed.

The added patch fixes the problem in the upstream sources; one of the
existing patches has to be adjusted to still apply.

Signed-off-by: Daniel Klauer <daniel.klauer at gin.de>
---
 .../python/python-smartpm/smart-attempt.patch      |  6 +--
 .../smart-rpm-transaction-failure-check.patch      | 57
++++++++++++++++++++++
 meta/recipes-devtools/python/python-smartpm_git.bb |  1 +
 3 files changed, 61 insertions(+), 3 deletions(-)  create mode 100644
meta/recipes-devtools/python/python-smartpm/smart-rpm-transaction-failure-ch
eck.patch

diff --git a/meta/recipes-devtools/python/python-smartpm/smart-attempt.patch
b/meta/recipes-devtools/python/python-smartpm/smart-attempt.patch
index ec98e03..5aedc88 100644
--- a/meta/recipes-devtools/python/python-smartpm/smart-attempt.patch
+++ b/meta/recipes-devtools/python/python-smartpm/smart-attempt.patch
@@ -36,7 +36,7 @@ index 9bbd952..ba6405a 100644
          finally:
              del getTS.ts
              cb.grabOutput(False)
-+            if probs and sysconf.has("attempt-install", soft=True):
++            if (probs is not None) and sysconf.has("attempt-install",
soft=True):
 +                def remove_conflict(pkgNEVR):
 +                    for key in changeset.keys():
 +                        if pkgNEVR == str(key):
@@ -67,8 +67,8 @@ index 9bbd952..ba6405a 100644
 +                        retry = 0
 +
              prog.setDone()
--            if probs:
-+            if probs and (not retry):
+-            if probs is not None:
++            if (probs is not None) and (not retry):
                  raise Error, "\n".join([x[0] for x in probs])
              prog.stop()
 +            if retry and len(changeset):
diff --git
a/meta/recipes-devtools/python/python-smartpm/smart-rpm-transaction-failure-
check.patch
b/meta/recipes-devtools/python/python-smartpm/smart-rpm-transaction-failure-
check.patch
new file mode 100644
index 0000000..a740ddd
--- /dev/null
+++ b/meta/recipes-devtools/python/python-smartpm/smart-rpm-transaction-
+++ failure-check.patch
@@ -0,0 +1,57 @@
+From 0c55d7e18f40465e95e8e4bf22af01f5d4477d3c Mon Sep 17 00:00:00 2001
+From: Daniel Klauer <daniel.klauer at gin.de>
+Date: Wed, 11 May 2016 17:22:49 +0200
+Subject: [PATCH] rpm: Don't ignore transaction error with empty
+problems list
+
+SmartPM could misinterpret RPM transaction error as success, if
+ts.run() (RPM Python API) returns an empty problems list, because of
+incorrect check for None which treated empty list to be the same as
+None when it has different meaning.
+
+ts.run() returns:
+* None in case of success
+* problems list in case of error, may be empty (look at rpmts_Run() in
+rpm-5.4.14/python/rpmts-py.c [1])
+
+"if mylist" is not good enough to check for error here, because it will
+treat an empty list as "false" because its len() == 0 [2].
+
+ts.check() seems to be different (it's ok for it to return an empty
+list), but for consistency it should be made clear that it can return
+either None, an empty list or a non-empty list.
+
+[1] http://rpm5.org/cvs/fileview?f=rpm/python/rpmts-py.c&v=1.111.2.3
+[2] https://docs.python.org/2/library/stdtypes.html#truth-value-testing
+
+Upstream-Status: Pending
+
+Signed-off-by: Daniel Klauer <daniel.klauer at gin.de>
+---
+ smart/backends/rpm/pm.py | 4 ++--
+ 1 file changed, 2 insertions(+), 2 deletions(-)
+
+diff --git a/smart/backends/rpm/pm.py b/smart/backends/rpm/pm.py index
+9bbd952..635f726 100644
+--- a/smart/backends/rpm/pm.py
++++ b/smart/backends/rpm/pm.py
+@@ -208,7 +208,7 @@ class RPMPackageManager(PackageManager):
+         force = sysconf.get("rpm-force", False)
+         if not force:
+             probs = ts.check()
+-            if probs:
++            if (probs is not None) and (len(probs) != 0):
+                 problines = []
+                 for prob in probs:
+                     name1 = "%s-%s-%s" % prob[0] @@ -247,7 +247,7 @@
+class RPMPackageManager(PackageManager):
+             del getTS.ts
+             cb.grabOutput(False)
+             prog.setDone()
+-            if probs:
++            if probs is not None:
+                 raise Error, "\n".join([x[0] for x in probs])
+             prog.stop()
+
+--
+1.9.1
+
diff --git a/meta/recipes-devtools/python/python-smartpm_git.bb
b/meta/recipes-devtools/python/python-smartpm_git.bb
index d9a908d..0d7f88c 100644
--- a/meta/recipes-devtools/python/python-smartpm_git.bb
+++ b/meta/recipes-devtools/python/python-smartpm_git.bb
@@ -17,6 +17,7 @@ SRC_URI = "\
           file://smart-recommends.patch \
           file://smart-improve-error-reporting.patch \
           file://smart-channelsdir.patch \
+          file://smart-rpm-transaction-failure-check.patch \
           file://smart-attempt.patch \
           file://smart-attempt-fix.patch \
           file://smart-add-for-rpm-ignoresize-check.patch \
--
1.9.1

--
_______________________________________________
Openembedded-core mailing list
Openembedded-core at lists.openembedded.org
http://lists.openembedded.org/mailman/listinfo/openembedded-core







More information about the Openembedded-core mailing list