Firstly, Javadoc 20 cannot handle source level 1.7 anymore, so I bumped
it to 11, which hopefully will be around for another while. It also
matches the ajc minimum of Java 11, i.e. ajdoc and ajc are synchronised
in this regard now.
Secondly, as of JDK 20, the javadoc tool renders HTML pages with section
headers specific to the JVM default locale. The workaround is to set the
locale to English before calling javadoc.
Signed-off-by: Alexander Kriegisch <Alexander@Kriegisch.name>
Improve Java 9+ version matching in ClassPath.getClassPath
On the short term, this fixes the failing weaver tests on JDK 20. On the
long term, the improved matching heuristics should make it unnecessary
to update the regex again.
Signed-off-by: Alexander Kriegisch <Alexander@Kriegisch.name>
Let's find out if the tests still pass
- on Java 19, except for J19 preview features,
- on Java 20, also except for J19 with changed syntax.
TODO: After having added Java 20 preview feature tests, remove J19
build.
Signed-off-by: Alexander Kriegisch <Alexander@Kriegisch.name>
Let's find out if the tests still pass
- on Java 19, except for J19 preview features,
- on Java 20, also except for J19 with changed syntax.
Signed-off-by: Alexander Kriegisch <Alexander@Kriegisch.name>
The test started breaking on GitHub, but ran fine locally, which implied
some kind of classpath ordering issue. As it turned out, for XML tests
all JARs found in the sandbox directory are added to the classpath
automatically. So if we do not want them on the classpath, we need to
delete or rename them before running.
In this case, however, it was enough to make sure the classpath order is
correct, so that the duplicate 'Application' class is found in the woven
version, not in the unwoven one by chance. We therefore need to use the
correct syntax,
1. separating classpath entries by comma, not space,
2. prepending '$sandbox/' to the JAR name.
Now the test reliably does what it should have done from the start: run
the woven code. Until now, the output assertions were adjusted to the
weird result of no aspect logs being present on the console, which was
wrong right from the start.
Signed-off-by: Alexander Kriegisch <Alexander@Kriegisch.name>
I'm the author of mill-aspectj and Mill maintainer, so i might be biased. ;-)
We use Mill with the mill-aspectj plugin in commercial projects, so I'm rather confident in it's reliability.
Further improve indentation of compilation results on the console
In CompilationResult.toString, no longer indent like this:
[warning 1] warning at
after() : execution(FooBar Blah.*()) {
^^^^^^
xxx FooBar [Xlint:invalidAbsoluteTypeName]
Instead, always just indent by 2 spaces, saving screen real estate:
[warning 1] warning at
after() : execution(FooBar Blah.*()) {
^^^^^^
xxx FooBar [Xlint:invalidAbsoluteTypeName]
Also further streamline/simplify the code a bit.
Signed-off-by: Alexander Kriegisch <Alexander@Kriegisch.name>
Fix indentation of compilation results on the console
Whenever warnings or errors were printed via CompilationResult.toString,
indirectly also using MessageUtil.renderMessage(IMessage, boolean),
messages containing context info such as code snippets with carets
marking erroneous tokens - see also the previous commit - prefixes like
"[warning 1] warning at " were printed right in front of the code
snippets. I.e., the carets marking erroneous tokens in the second line
were not indented like the first line with the code snippet, leading to
(simplified) output like:
[warning 1] warning at after() : execution(FooBar Blah.*()) {
^^^^^^
xxx FooBar [Xlint:invalidAbsoluteTypeName]
This was fixed to now correctly indent lines 2 to n according to line 1,
yielding the correct output:
[warning 1] warning at
after() : execution(FooBar Blah.*()) {
^^^^^^
xxx FooBar [Xlint:invalidAbsoluteTypeName]
Especially with longer, more complex context lines, this helps to
identify the erroneous section. BTW, for one-line messages, everything
of course looks like before.
Signed-off-by: Alexander Kriegisch <Alexander@Kriegisch.name>
Bugfix: caret error marks in compiler output too short
This very old bug in EclipseAdapterUtils calculated the '^' caret error
marks incorrectly. The marks were too short like this:
void m() { return vax[3]; }
^^^^^^^^^^^
Where the correct result would look like this:
void m() { return vax[3]; }
^^^^^^^^^^^^^^
This was due to the rather complicated way in which first surrounding
context code - here the leading 'void m() { ' and trailing ' }' - was
first added and then possible leading white space was cut off again from
the left. It is difficult to explain, the code is a nightmare, trying to
work with char arrays, counting indexes, repeatedly using
System.arraycopy and using lots of counters and offsets. I would have
liked to simplify the code, converting char[] buffers to Strings, but
decided to keep the basic structure for now, not sure what kind of
memory or performance considerations led to this design.
Signed-off-by: Alexander Kriegisch <Alexander@Kriegisch.name>
Comment on newly pointcut parsing found problem in test class
A pointcut like
execution(*..Collection<?>[] *(..))
leads to an AJ core dump, which is something I noticed while fixing a
test in the previous commit. I am going to create a new issue for it and
link to this commit later.
Signed-off-by: Alexander Kriegisch <Alexander@Kriegisch.name>
After the changes, the weaver's behaviour is now slightly different, but
not actually buggy. Actually, in one case there is now a weaver warning
for a non-matching pointcut which should have been there before, but was
not. I.e., things have improved and the tests are even a bit better now.
Signed-off-by: Alexander Kriegisch <Alexander@Kriegisch.name>
Fix PointcutRewriterTest, add LogicalPointcutStructure test helper class
After WildTypePattern.hashCode was fixed in the previous commit,
PointcutRewriterTest started failing, because in many places it was
falsely relying on a specific order of hash codes, which cannot be
guaranteed, especially since more instance fields are part of the hash
code now in accordance with 'equals'.
The new test helper class LogicalPointcutStructure is able to recognise
chained '&&' and '||' pointcuts of the same logical nesting level,
un-nesting them from the actual pointcut structure and making them
comparable, disregarding their order. I.e., something like
((A && B) && C) && D
is actually recognised to logically be
A && B && C && D
and equivalent to e.g. either of
D && B && A && C
A && B && D && C
C && A && D && B
This helps to compare rewritten pointcuts, as long as their logical
structure has not been altered.
Relates to #24.
Signed-off-by: Alexander Kriegisch <Alexander@Kriegisch.name>
WildTypePattern: fix hashCode and toString methods
Especially 'hashCode' did not correspond to 'equals', disregarding
several fields, array dimension information being only one of them. This
led to parts of pointcuts being ignored, because they were regarded as
duplicates. Example:
execution(Foo* *(..)) && !execution(Foo*[] *(..))
Here, the negated pattern was falsely regarded as equal to the first
pattern, leading to an "A && !A" situation, i.e. no match at all.
Furthermore, 'toString' did not print array strings, i.e. instead of
"Foo*[][]" something like "Foo*" was printed. This false information was
also present in annotations generated by the weaver.
FuzzilyMatchingAspect was adjusted to actually match exactly once, as
expected, for the "Foo*" return types, i.e. exclusions for the array
return types have been added.
Relates to #24.
Signed-off-by: Alexander Kriegisch <Alexander@Kriegisch.name>
Handle one- and multi-dimensional array return types correctly
Fixes https://github.com/eclipse/org.aspectj/issues/24, both the array
return type matching as such as well as matching dimensionality patterns
correctly. E.g., 'Foo*[]' is not the same as 'Foo*[][]'. This also works
correctly in combination with asterisks, even for primitive types, i.e.
'in*[][]' correctly matches a 2-dimensional array of 'int'.
Signed-off-by: Alexander Kriegisch <Alexander@Kriegisch.name>
Simplify if-else in WildTypePattern.matchesExactlyByName
A simple boolean condition is enough.
Loosely relates to https://github.com/eclipse/org.aspectj/issues/24, but
actually it is just drive-by cosmetics.
Signed-off-by: Alexander Kriegisch <Alexander@Kriegisch.name>
The method falsely determined that a one-dimensional array was not an
array due to a one-off bug.
Relates to https://github.com/eclipse/org.aspectj/issues/24.
Signed-off-by: Alexander Kriegisch <Alexander@Kriegisch.name>
More test cases were added for
- multi-dimensional arrays,
- primitive type arrays.
Relates to https://github.com/eclipse/org.aspectj/issues/24.
Signed-off-by: Alexander Kriegisch <Alexander@Kriegisch.name>
Refactor, comment and reactivate Ajc183Tests.testAnnoStyleDecp_442425
This test was always inactive and did not add any value other than
during development. No it runs, passes and documents the status quo of
- what was fixed (regression for AJC 1.8.2 core dump fixed in 1.8.3),
- the current limitations of @DeclareParents and @DeclareMixin
regarding generic interfaces.
Signed-off-by: Alexander Kriegisch <Alexander@Kriegisch.name>
BcelTypeMunger.mungeMethodDelegate: only use 'synchronized' when necessary
Relates to #198. Now, we create a delegate method body which basically
looks as follows:
public void methodOne() {
if (this.ajc$instance$MyAspect$MyMixin == null) {
synchronized(this) {
if (this.ajc$instance$MyAspect$MyMixin == null) {
this.ajc$instance$MyAspect$MyMixin = MyAspect.aspectOf().createImplementation(this);
}
}
}
this.ajc$instance$MyAspect$MyMixin.methodOne();
}
The idea for the outer null check is from @aclement, see
https://github.com/eclipse/org.aspectj/pull/205#issuecomment-1371556080.
Signed-off-by: Alexander Kriegisch <Alexander@Kriegisch.name>
No more compiler errors for implicitly static inner aspects of interfaces
Fixes #162. Contains regression test
Bugs1919Tests.testInterfaceInnerAspectImplicitlyStatic.
Signed-off-by: Alexander Kriegisch <Alexander@Kriegisch.name>
Improve BcelTypeMunger.mungeMethodDelegate to avoid race condition
Fixes #198, i.e. test DeclareMixinTests.testCaseEConcurrent from the
previous commit now passes.
Signed-off-by: Alexander Kriegisch <Alexander@Kriegisch.name>
Alexander Kriegisch
Tobias Roeser