fix(main/openjdk-{17,21,25}): fix multiple crashes

[robertkirkman]

• Fixes [Bug]: 32-bit ARM build of openjdk-25 has Illegal instruction (also, unpatched TMP_BUFFER_LEN discovered) #27045

• Disables g1gc and LTO for 32-bit ARM

  • Partially reverts openjdk/jdk24u@0b467e9
  • Partially reverts openjdk/jdk24u@85fedbf
  • Fixes [Bug]: 32-bit ARM build of openjdk-25 has Illegal instruction (also, unpatched TMP_BUFFER_LEN discovered) #27045

• It is also necessary to hardcode char* tmpdir = "@TERMUX_PREFIX@/tmp"; because OpenJDK has an internal hostbuild that will crash at build-time if it is not also redirected to use @TERMUX_PREFIX@/tmp instead of /tmp simultaneously

  • The crash is only easily reproducible in the 32-bit ARM build of openjdk-25, but the same code exists possibly-dormant in every OpenJDK, so should be patched in all
  • This is OK because @TERMUX_PREFIX@/tmp is already a directory excluded from packaging globally, so it is safe to use as a tmpdir at build-time of anything

termux-packages/scripts/build/termux_step_copy_into_massagedir.sh

Line 5 in f7d3769

tar -C "$TERMUX_PREFIX_CLASSICAL" -N "$TERMUX_BUILD_TS_FILE" --exclude='tmp' -cf - . | \

[robertkirkman]

more changes may be necessary, please wait

[robertkirkman]

Status update:

• After 30 hours of bisecting, 2 commits which break OpenJDK on 32-bit ARM Android have been found in between OpenJDK 23 and OpenJDK 24 and successful workarounds for them written and applied here to OpenJDK 25.

• It is believed that there are at least 2 other commits which also break OpenJDK on 32-bit ARM Android in other ways that are not yet found.

  • One of the commits is believed to be introduced somewhere in between OpenJDK 17 and OpenJDK 21
    • and there is also believed to be a follow-up commit that unbreaks it somewhere in between OpenJDK 21 and OpenJDK 22
  • The other commit is believed to be introduced somewhere in between OpenJDK 21 and OpenJDK 22
    • (and current bisecting progress indicates that it seems to be somewhere shortly after, not before or the same as, the commit that unbreaks the commit that broke OpenJDK 21)

I am currently building and testing hundreds of intermediate development commits of OpenJDK for 32-bit ARM Android in order to bisect the remaining two commits that are currently unknown, working backwards from newer to older OpenJDKs until all commits broken on 32-bit ARM Android are precisely identified and then reverted or worked around.