Robert Schuster writes:
> Hi guys,
> today I tripped on some weird incompatibility between ecj (3.3.1 as
> found in Debian Lenny) and gjar 0.95:
>
> gjar creates jars which need zip version 2.0 to extract, while e.g.
> fastjar creates files which are readable by version 1.0.
>
> This is what 'file' says:
>
> src1-fastjar.jar: Zip archive data, at least v1.0 to extract
> src1-gjar.jar: Zip archive data, at least v2.0 to extract
>
> The problem comes when you want to compile something with ecj and put
> such a jar (the gjar-created one) in the classpath or bootclasspath.
>
> It simply does not find the files!
>
> ecj -cp ../src1-gjar.jar use/Test.java
Now, this is very odd. It really shouldn't happen unless the runtime
libraries used by ecj and by jar are different.
> ----------
> 1. ERROR in use/Test.java (at line 3)
> import base.Base;
> ^^^^
> The import base cannot be resolved
> ----------
> 2. ERROR in use/Test.java (at line 9)
> Base b = new Base("Test");
> ^^^^
> Base cannot be resolved to a type
> ----------
> 3. ERROR in use/Test.java (at line 9)
> Base b = new Base("Test");
> ^^^^
> Base cannot be resolved to a type
> ----------
> 3 problems (3 errors)rob@linkist:~/tmp/jartest/src2$
>
> Looking at the jar's content reveals that a class is there:
> gjar -tf src1-gjar.jar
>
> META-INF/MANIFEST.MF
> ./base/Base.class
>
> Who is the one to blame? Is ecj unable to process version 2.0 zip files
> or does gjar create unsuitable files? Is this a known issue?
Sun's jar also sometimes creates zip v2.0 files. Try this:
touch file.file
{jdk}/bin/jar cf x.jar file.file
x.jar: Zip archive data, at least v2.0 to extract
ecj uses the runtime library's java.util.zip.*, so it can't simply be
the zip version that's causing the problem?
I wonder: is it simply that some files are compressed? The DEFLATE
algorithm is probably why unzip version 2.0 is needed. The difference
is that some files are compressed.
unzip -lv ../src1-gjar.jar
Archive: ../src1-gjar.jar
Length Method Size Ratio Date Time CRC-32 Name
-------- ------ ------- ----- ---- ---- ------ ----
25 Defl:N 27 -8% 12-03-07 15:30 ee027fb2 META-INF/MANIFEST.MF
354 Stored 354 0% 12-03-07 15:30 0bee6f8e ./base/Base.class
-------- ------- --- -------
379 381 -1% 2 files
unzip -lv ../src1-fastjar.jar
Archive: ../src1-fastjar.jar
Length Method Size Ratio Date Time CRC-32 Name
-------- ------ ------- ----- ---- ---- ------ ----
0 Stored 0 0% 12-03-07 15:28 00000000 META-INF/
44 Stored 44 0% 12-03-07 15:28 616d2db8 META-INF/MANIFEST.MF
354 Stored 354 0% 11-29-07 23:42 0bee6f8e base/Base.class
-------- ------- --- -------
398 398 0% 3 files
unzip -lv ../src1-jar.jar
Archive: ../src1-jar.jar
Length Method Size Ratio Date Time CRC-32 Name
-------- ------ ------- ----- ---- ---- ------ ----
0 Stored 0 0% 12-03-07 15:33 00000000 META-INF/
71 Stored 71 0% 12-03-07 15:33 8f35f6c2 META-INF/MANIFEST.MF
354 Stored 354 0% 11-29-07 23:42 0bee6f8e base/Base.class
-------- ------- --- -------
425 425 0% 3 files
OK, I'm going to do an experiment. Let's try this:
$ gjar cvf ../src1-gjar.jar ./base/Base.class
adding: ./base/Base.class (in=354) (out=354) (stored 26%)
$ jar cvf ../src1-gjar.jar ./base/Base.class
added manifest
adding: base/Base.class(in = 354) (out= 265)(deflated 25%)
Note that gjar adds "./" to the start of the path, and jar does not.
That's the bug: gjar needs to strip "./" from the start of the path.
$ gjar cvf ../src1-gjar.jar base/Base.class
adding: base/Base.class (in=354) (out=354) (stored 26%)
$ cd ../src2/
$ ./compile-gjar
----------
1. WARNING in use/Test.java (at line 9)
Base b = new Base("Test");
^
The local variable b is never read
----------
1 problem (1 warning)
Andrew.
--
Red Hat UK Ltd, Amberley Place, 107-111 Peascod Street, Windsor, Berkshire, SL4 1TE, UK
Registered in England and Wales No. 3798903
