[Scons-users] more Java scons3 issues...

[Mats Wichmann]

I'm running into this:

Code snippet:

jdk_env.Java(target='classes', source=['src/main/java'])
example_jar = jdk_env.Jar(target='simpleclientserver.jar',
source=['classes', 'MANIFEST.MF'],
JARCHDIR='$SOURCE')

When run on 2.5.1, as expected (since the above stanza was adjusted
until it worked - 'classes' is in the source instead of the built
classes due to the Java() problem that it can't always reliably detect
what javac will emit, the case described in
http://scons.tigris.org/issues/show_bug.cgi?id=2547 being in play here)

jar cfm
out/linux/x86_64/debug/java/examples-java/simpleclientserver/simpleclientserver.jar
java/examples-java/simpleclientserver/MANIFEST.MF -C
out/linux/x86_64/debug/java/examples-java/simpleclientserver/classes .


When run on 3.0.1:

jar cfm
out/linux/x86_64/debug/java/examples-java/simpleclientserver/simpleclientserver.jar
java/examples-java/simpleclientserver/MANIFEST.MF -C
out/linux/x86_64/debug/java/examples-java/simpleclientserver/classes .
-C out/linux/x86_64/debug/java/examples-java/simpleclientserver/classes
org -C
out/linux/x86_64/debug/java/examples-java/simpleclientserver/classes
org/iotivity -C
out/linux/x86_64/debug/java/examples-java/simpleclientserver/classes
org/iotivity/base -C
out/linux/x86_64/debug/java/examples-java/simpleclientserver/classes
org/iotivity/base/examples
java.util.zip.ZipException: duplicate entry: org/


After some experimentation, it looks like this problem is taken care of
if I drop in the latest Tool/jar.py from github:

jar cf
out/linux/x86_64/debug/java/examples-java/simpleclientserver/simpleclientserver.jar
-C out/linux/x86_64/debug/java/examples-java/simpleclientserver/classes .


but... now the manifest isn't being included.

ideas?

[Bill Deegan]

Mats,

I see you're filing a bug for this.
If you check the bug filing policy we ask that you wait for someone to
approve filing a new bug to prevent duplicates, etc from filling up the bug
tracker.

Have you tried the lastest changes in the master branch?
There are some new Java fixes in there since 3.0.1..

-Bill

Post by Mats Wichmann
jdk_env.Java(target='classes', source=['src/main/java'])
example_jar = jdk_env.Jar(target='simpleclientserver.jar',
source=['classes', 'MANIFEST.MF'],
JARCHDIR='$SOURCE')
When run on 2.5.1, as expected (since the above stanza was adjusted
until it worked - 'classes' is in the source instead of the built
classes due to the Java() problem that it can't always reliably detect
what javac will emit, the case described in
http://scons.tigris.org/issues/show_bug.cgi?id=2547 being in play here)
jar cfm
out/linux/x86_64/debug/java/examples-java/simpleclientserver/
simpleclientserver.jar
java/examples-java/simpleclientserver/MANIFEST.MF -C
out/linux/x86_64/debug/java/examples-java/simpleclientserver/classes .
jar cfm
out/linux/x86_64/debug/java/examples-java/simpleclientserver/
simpleclientserver.jar
java/examples-java/simpleclientserver/MANIFEST.MF -C
out/linux/x86_64/debug/java/examples-java/simpleclientserver/classes .
-C out/linux/x86_64/debug/java/examples-java/simpleclientserver/classes
org -C
out/linux/x86_64/debug/java/examples-java/simpleclientserver/classes
org/iotivity -C
out/linux/x86_64/debug/java/examples-java/simpleclientserver/classes
org/iotivity/base -C
out/linux/x86_64/debug/java/examples-java/simpleclientserver/classes
org/iotivity/base/examples
java.util.zip.ZipException: duplicate entry: org/
After some experimentation, it looks like this problem is taken care of
jar cf
out/linux/x86_64/debug/java/examples-java/simpleclientserver/
simpleclientserver.jar
-C out/linux/x86_64/debug/java/examples-java/simpleclientserver/classes .
but... now the manifest isn't being included.
ideas?
_______________________________________________
Scons-users mailing list
https://pairlist4.pair.net/mailman/listinfo/scons-users

[Mats Wichmann]

Post by Bill Deegan
Mats,
I see you're filing a bug for this.
If you check the bug filing policy we ask that you wait for someone to
approve filing a new bug to prevent duplicates, etc from filling up the bug
tracker.
Have you tried the lastest changes in the master branch?
There are some new Java fixes in there since 3.0.1..

I've checked for duplication, but this is effectively new code. pre-3.0
code in jar.py did not do anything like this. The post 3.0.1 jar.py
changes help other problems I have, but not this one.

The problem is in a simple check, os.path.isfile(), which is not safe to
do in the context of a variant dir (without copying), because the source
files in the working/variant directory exist only in the mind of SCons,
not Python. As noted in the bug, I don't at the moment know how to apply
a transformation to convert back to the source directory in the context
in which the tool runs, so figured it was best to document it.

http://scons.tigris.org/issues/show_bug.cgi?id=3042

[Bill Deegan]

Did you try the current head of the master branch?

Post by Mats Wichmann

Post by Bill Deegan
Mats,
I see you're filing a bug for this.
If you check the bug filing policy we ask that you wait for someone to
approve filing a new bug to prevent duplicates, etc from filling up the

bug

Post by Bill Deegan
tracker.
Have you tried the lastest changes in the master branch?
There are some new Java fixes in there since 3.0.1..

I've checked for duplication, but this is effectively new code. pre-3.0
code in jar.py did not do anything like this. The post 3.0.1 jar.py
changes help other problems I have, but not this one.
The problem is in a simple check, os.path.isfile(), which is not safe to
do in the context of a variant dir (without copying), because the source
files in the working/variant directory exist only in the mind of SCons,
not Python. As noted in the bug, I don't at the moment know how to apply
a transformation to convert back to the source directory in the context
in which the tool runs, so figured it was best to document it.
http://scons.tigris.org/issues/show_bug.cgi?id=3042
_______________________________________________
Scons-users mailing list
https://pairlist4.pair.net/mailman/listinfo/scons-users

[Mats Wichmann]

Post by Bill Deegan
Did you try the current head of the master branch?

Post by Mats Wichmann
The post 3.0.1 jar.py
changes help other problems I have, but not this one.

[Daniel Moody]

Hey Mats,

I am looking into this issue, where does the MANIFEST.MF file exist at the
start of the build relative to the Scons file you mentioned above?

What is your VariantDir call look like?

Best regards

Post by Mats Wichmann

Post by Bill Deegan
Did you try the current head of the master branch?

Post by Mats Wichmann
The post 3.0.1 jar.py
changes help other problems I have, but not this one.

_______________________________________________
Scons-users mailing list
https://pairlist4.pair.net/mailman/listinfo/scons-users

[Daniel Moody]

Can you check out the changes from this pull request and see if it fixes
your issue?

https://github.com/SCons/scons/pull/35

Post by Daniel Moody
Hey Mats,
I am looking into this issue, where does the MANIFEST.MF file exist at the
start of the build relative to the Scons file you mentioned above?
What is your VariantDir call look like?
Best regards

Post by Mats Wichmann

Post by Bill Deegan
Did you try the current head of the master branch?

Post by Mats Wichmann
The post 3.0.1 jar.py
changes help other problems I have, but not this one.

_______________________________________________
Scons-users mailing list
https://pairlist4.pair.net/mailman/listinfo/scons-users

[Mats Wichmann]

Post by Daniel Moody
Can you check out the changes from this pull request and see if it fixes
your issue?
https://github.com/SCons/scons/pull/35

I will try this presently. I was exploring something down this path
earlier with the same thinking. My experiments indicated that the
conversion to File fails only if the string refers to an existing
directory (i.e. you still get a node even if the string does not refer
to an existing file), so while it looks likely to work for our case of
having a manifest file plus already-generated classfiles, it would
prevent the code from hitting the later part of the loop - lines 206 and
later in your patched version. I don't have proof of that conjecture at
this point, and will update here when I get a chance to test.

Post by Daniel Moody

Post by Daniel Moody
Hey Mats,
I am looking into this issue, where does the MANIFEST.MF file exist at the
start of the build relative to the Scons file you mentioned above?
What is your VariantDir call look like?
Best regards

Post by Mats Wichmann

Post by Bill Deegan
Did you try the current head of the master branch?

Post by Mats Wichmann
The post 3.0.1 jar.py
changes help other problems I have, but not this one.

_______________________________________________
Scons-users mailing list
https://pairlist4.pair.net/mailman/listinfo/scons-users

_______________________________________________
Scons-users mailing list
https://pairlist4.pair.net/mailman/listinfo/scons-users

[Daniel Moody]

Hey mats,

I think you might be right, actually the last part was to catch the nodes
that passed through python isfile or isdir checks previously, so now it may
not be needed.

There are cases where it gets passed the two trys added in the PR which I
can tell from the swig tests which hit the ".i" extension check, but I am
going to investigate the ordering of checks in that PR because it doesn't
seem like it should be able to get passed the two tries.

Best regards

Post by Mats Wichmann

Post by Daniel Moody
Can you check out the changes from this pull request and see if it fixes
your issue?
https://github.com/SCons/scons/pull/35

I will try this presently. I was exploring something down this path
earlier with the same thinking. My experiments indicated that the
conversion to File fails only if the string refers to an existing
directory (i.e. you still get a node even if the string does not refer
to an existing file), so while it looks likely to work for our case of
having a manifest file plus already-generated classfiles, it would
prevent the code from hitting the later part of the loop - lines 206 and
later in your patched version. I don't have proof of that conjecture at
this point, and will update here when I get a chance to test.

Post by Daniel Moody

Post by Daniel Moody
Hey Mats,
I am looking into this issue, where does the MANIFEST.MF file exist at

the

Post by Daniel Moody

Post by Daniel Moody
start of the build relative to the Scons file you mentioned above?
What is your VariantDir call look like?
Best regards

Post by Mats Wichmann

Post by Bill Deegan
Did you try the current head of the master branch?

Post by Mats Wichmann
The post 3.0.1 jar.py
changes help other problems I have, but not this one.

_______________________________________________
Scons-users mailing list
https://pairlist4.pair.net/mailman/listinfo/scons-users

_______________________________________________
Scons-users mailing list
https://pairlist4.pair.net/mailman/listinfo/scons-users

_______________________________________________
Scons-users mailing list
https://pairlist4.pair.net/mailman/listinfo/scons-users