On Tue, 21 Jan 2025 10:26:41 GMT, Kevin Walls <kev...@openjdk.org> wrote:
>> This fix widens the regex used to determine if a JAVAHEAP line is correct in >> the output of System.map/System.dump_map. > > Hi, > From our logs - existing passes on macos aarch64 include a line which > matches, e.g.: > > 0x0007c8000000-0x0007d0000000 134217728 rw-/rwx pvt 0 > JAVAHEAP > > The failures where we don't match, they have p/a rather than pvt: > > 0x00030e900000-0x00030ec00000 3145728 rw-/rwx p/a 0 JAVAHEAP > > There are also the "cow" lines, I see in the JAVAHEAP classes.jsa line. > > I think we don't specifically need "cow" in the macprivate_or_shared > definition - as long as we match the JAVAHEAP line with p/a we will pass. > It's only classes.jsa marked a copy on write (makes sense), and there should > always be another JAVAHEAP line for us to match. > > Presuming "private aliased" is a variation of private, so adding p/a variant > is a trivial change for the test. @kevinjwalls thanks for the suggestion. I wish I could duplicate your issue. I prefer to leave 'cow' in, as the goal of this test is to ensure System.map works, not to confirm the implementation of a specific garbage collector. That was the consensus reached when ZGC caused the first failure on macOS. ------------- PR Comment: https://git.openjdk.org/jdk/pull/23205#issuecomment-2604721965
