[Scons-users] Bug causing corruption of .sconsign.dblite

Mon Oct 20 10:20:20 EDT 2014

Stijn,

Negative.  The issue has nothing to do with GIT.

The bug exists within the "MD5-timestamp" decider.  This decider can
falsely mark Nodes as "not up-to-date" because of how the dependency cache
is expanded when new dependencies are added to a Node.  In worst case,
sometimes code will get compiled that wasn't necessary, but only after the
timestamp on those files would change; thus, it's a performance bug, but in
most cases will still be faster than using the "MD5" decider since it
doesn't need to compute the MD5 sums every pass.

V/R,
William

On Mon, Oct 20, 2014 at 4:31 AM, Stijn De Ruyck <Stijn.DeRuyck at onsemi.com>
wrote:

>  I’m trying to understand this issue. Can someone summarize?
>
>
>
> Does it mean Scons 2.3.0 is essentially unusable when combined with git?
>
> I’m currently converting 2 software projects to Scons, but I’m stuck with
> 2.3.0 because of Python dependencies…
>
>
>
> Any checkout, pull, rebase, reset, … can corrupt the .sconsign.dblite file?
>
>
>
> Best regards,
>
> Stijn
>
>
>
>
>
> *From:* Scons-users [mailto:scons-users-bounces at scons.org] *On Behalf Of *William
> Blevins
> *Sent:* Sunday, October 19, 2014 9:21 PM
> *To:* SCons users mailing list
> *Subject:* Re: [Scons-users] Bug causing corruption of .sconsign.dblite
>
>
>
> Piotr,
>
>
>
> I applied the patch to the latest SCons source, and it broke several
> automated tests.  I do not know whether those tests simply need to be
> updated or additional modifications are required.
>
>
>
> I am currently working on another issue regarding cross-language scanner
> support, and I cannot dedicate the time to investigate at present.  If you
> are eager to have this fixed in the next SCons update, consider forking a
> repository @ https://bitbucket.org/scons/scons and creating a pull
> request.  Additional developer notes can be found @
> http://www.scons.org/wiki and there are individuals who can help if you
> have additional questions.
>
>
>
> V/R,
>
> William
>
>
>
> On Fri, Oct 17, 2014 at 8:37 PM, William Blevins <wblevins001 at gmail.com>
> wrote:
>
> Piotr,
>
>
>
> A software failure indeed.  After your test steps, the md5sum for beta.h
> is incorrect.  If I update the timestamp for beta.h, main.cpp will
> incorrectly rebuild.
>
>
>
> Thank you for a streamlined test case. I made a issue tracker for this
> defect: http://scons.tigris.org/issues/show_bug.cgi?id=2980
>
>
>
> V/R,
>
> William
>
>
>
> On Thu, Oct 16, 2014 at 7:58 PM, William Blevins <wblevins001 at gmail.com>
> wrote:
>
> Piotr,
>
>
>
> Thanks for supplying the additional information. I will make my best
> effort to take a close look at the information this weekend.  In the
> unlikely event that a SCons developer hasn't gotten back to you by end of
> next week, please ping us with a gentle reminder.
>
>
>
> V/R,
>
> William
>
>
>
> On Thu, Oct 16, 2014 at 10:40 AM, Piotr Bartosik <piotr.bartosik at gmail.com>
> wrote:
>
> William,
>
> please find attached a simple test that reproduces the bug. I tested it on
> Linux. Below is my example output with discrepancies highlighted:
>
> $ ./test.sh ~/tools/scons
> Note: checking out '6ee27c0db68515e3faa71bf25c0ed7a883ecf697'.
>
> You are in 'detached HEAD' state. You can look around, make experimental
> changes and commit them, and you can discard any commits you make in this
> state without impacting any branches by performing another checkout.
>
> If you want to create a new branch to retain commits you create, you may
> do so (now or later) by using -b with the checkout command again. Example:
>
>   git checkout -b new_branch_name
>
> HEAD is now at 6ee27c0... initial
> scons: Reading SConscript files ...
> scons: done reading SConscript files.
> scons: Building targets ...
> g++ -o main.o -c main.cpp
> g++ -o main main.o
> scons: done building targets.
> Previous HEAD position was 6ee27c0... initial
> Switched to branch 'master'
> scons: Reading SConscript files ...
> scons: done reading SConscript files.
> scons: Building targets ...
> g++ -o main.o -c main.cpp
> scons: done building targets.
> === .:
> SConstruct: None
> alpha.h: 3518d8c3310bba3c2a0da5489f569e14
> beta.h: 2ff783593a2224d0574181661ab5f1b7
> gamma.h: 2ff783593a2224d0574181661ab5f1b7
> main: b53741b91f13474487bbbc1efe5e328b
> main.cpp: ebf971b7da9c98f91067a9c601e96259
> main.o: faaa740a3704b2c8f034bafd7c1722ff
> === /usr/bin:
> g++: bb36ec5f9b8a8bfc02d1c4bf917a58e1
> 3518d8c3310bba3c2a0da5489f569e14  alpha.h
> bca4755df4cd2eef7d51bfb007c36dd1  beta.h
> 2ff783593a2224d0574181661ab5f1b7  gamma.h
> ebf971b7da9c98f91067a9c601e96259  main.cpp
>
>  Regards,
>
> Piotr
>
>
>
>
>
> 2014-10-15 16:00 GMT+02:00 William Blevins <wblevins001 at gmail.com>:
>
> Piotr,
>
>
>
> That should be fine.  Try to keep the dataset as simple and lightweight as
> possible, since a test case needs to be created from it.  If you only need
> a single c-file with a few header includes to reproduce, then that should
> be plenty.
>
>
>
> If you feel comfortable request a patch addition, you can do so here:
> https://bitbucket.org/scons/scons.
>
>
>
> V/R,
>
> William
>
>
>
> On Wed, Oct 15, 2014 at 5:13 AM, Piotr Bartosik <piotr.bartosik at gmail.com>
> wrote:
>
> William,
>
> thank you for your answer. I'm stating to prepare a test data set. I'm
> going to send you a compressed directory versioned with git (with .git dir
> inside) and an exact list of steps to reproduce the problem. I hope that
> this format is convenient to you, if not please let me know.
>
> Best regards,
> Piotr
>
>
>
> 2014-10-15 1:25 GMT+02:00 William Blevins <wblevins001 at gmail.com>:
>
> Piotr,
>
>
>
> Also, as a side question, do you have a set of test case values we can use
> for this case?
>
>
>
> V/R,
>
> William
>
>
>
> On Tue, Oct 14, 2014 at 7:23 PM, William Blevins <wblevins001 at gmail.com>
> wrote:
>
> Piotr,
>
>
>
> I am sure you already did your homework, and we appreciate your input.
> Before we start looking at this closely, can you guarantee that the
> following files were not accidentally committed your GIT repository: object
> files, libraries, executables, the .sconsign.dblite file itself?  I have
> seen cases in the past were files like these get committed accidentally,
> and that might cause effects like those you described.
>
>
>
> V/R,
>
> William
>
>
>
> On Tue, Oct 14, 2014 at 10:14 AM, Piotr Bartosik <piotr.bartosik at gmail.com>
> wrote:
>
>       Hi,
>
> while using SCons (version 2.3.0) together with git I encountered some
> problems with MD5 calculations. After rebasing to different revisions of my
> repository I noticed strange cache misses (the proper .o files should be
> already in the cache while they were actually rebuilt). This led me to the
> conclusion that some MD5 values may be calculated incorrectly. In fact,
> after using the script sconsign.py it became apparent that some MD5s in
> .sconsign.dblite were incorrect (the sums calculated directly using hashlib
> were different).
>
> To sum up, after this simple procedure:
>
> rm .sconsign.dblite
>
> git checkout SOME_VERSION_1
>
> scons
>
> git checkout SOME_VERSION_2
>
> scons
>
> sconsign.py reported incorrect MD5 sums.
>
> As it turned out, the problem is localized in Node/__init__.py. The code
>
> then.extend([None] * diff)
>
> in line 1050 causes desynchronization between two lists: the current
> dependencies and info nodes for dependencies from the previous build
> (containing, among others, previously calculated MD5 sums).
>
>
>
> For example, consider that in SOME_VERSION_1 a file has five dependencies:
>
> A B C D E
>
> and in SOME_VERSION_2 a new dependency appears:
>
> A B C X D E
>
> After line 1050, the lists "then" and "children" look like below:
>
> then:     A B C D E None
>
> children: A B C X D E
>
> Code further below updates children information. As it appears, such
> desynchronization does not cause any problems if we are checking MD5s each
> time: they just will be recalculated for X, D and E. But if we also look at
> the timestamps, the condition of timestamp remaining the same allows SCons
> to copy the MD5 sum from the "then" array to the current child object.
> Consequently, if D and E have the same modification date, E's checksum will
> be incorrectly copied into D's info. And this is exactly the error that I
> observe.
>
> My quick fix builds "then" list in a proper way, .i.e. without
> desynchronization (please see the attachment):
>
> then:     A B C None D E
>
> children: A B C X    D E
>
> Please look at it and tell me what you think. I don't know the SCons'
> sources very well so my fix may be a little naive but this is what I can
> provide for a start.
>
> Thanks,
>
> Piotr
>
>
>
> _______________________________________________
> Scons-users mailing list
> Scons-users at scons.org
> https://pairlist4.pair.net/mailman/listinfo/scons-users
>
>
>
>
>
>
> _______________________________________________
> Scons-users mailing list
> Scons-users at scons.org
> https://pairlist4.pair.net/mailman/listinfo/scons-users
>
>
>
>
> _______________________________________________
> Scons-users mailing list
> Scons-users at scons.org
> https://pairlist4.pair.net/mailman/listinfo/scons-users
>
>
>
>
> _______________________________________________
> Scons-users mailing list
> Scons-users at scons.org
> https://pairlist4.pair.net/mailman/listinfo/scons-users
>
>
>
>
> _______________________________________________
> Scons-users mailing list
> Scons-users at scons.org
> https://pairlist4.pair.net/mailman/listinfo/scons-users
>
>
>
>
>
>
>
> _______________________________________________
> Scons-users mailing list
> Scons-users at scons.org
> https://pairlist4.pair.net/mailman/listinfo/scons-users
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://pairlist4.pair.net/pipermail/scons-users/attachments/20141020/7ab73a51/attachment-0001.html>