Reproducible Build for v1.0.9 failed

[IzzySoft]

…with differences in the *.so files: diffoscope.html

As you do not use Github actions to build, I could not check what might have changed; my first idea was the NDK version (the ordering difference looks similar to what we saw here between NDK 27 and 29 with other apps), but I don't see any NDK involved in the build log. Any hints from your side? Thanks in advance!

[shiyunjin]

Because v1.0.8 enabled minifyEnabled and switched to the org.lsposed.hiddenapibypass:hiddenapibypass:6.1 package to use the SSL protocol, some devices were unable to pair due to the lack of the SSL extension. v1.0.9 rolled back this change and switched back to org.conscrypt:conscrypt-android:2.5.3.

e16a43f

This is the only change made in v1.0.9

[IzzySoft]

Matches the diff: it's only libconscrypt_jni.so, for each ABI, that differs:

  -rw-r--r--  0.0 unx  8114508 b-  2977180 defN 1981-01-01 01:01:02 e7aa1c80 classes.dex
- -rw-r--r--  0.0 unx  2103592 b-   818444 defN 1981-01-01 01:01:02 124e1de4 lib/arm64-v8a/libconscrypt_jni.so
+ -rw-r--r--  0.0 unx  2103592 b-   818443 defN 1981-01-01 01:01:02 0a771e01 lib/arm64-v8a/libconscrypt_jni.so
  -rw-r--r--  0.0 unx    28192 b-    18239 defN 1981-01-01 01:01:02 ed1bd724 lib/arm64-v8a/libspake2.so
- -rw-r--r--  0.0 unx  1284028 b-   637820 defN 1981-01-01 01:01:02 0da1130a lib/armeabi-v7a/libconscrypt_jni.so
+ -rw-r--r--  0.0 unx  1284028 b-   637822 defN 1981-01-01 01:01:02 bf9c0b92 lib/armeabi-v7a/libconscrypt_jni.so
  -rw-r--r--  0.0 unx    28888 b-    21621 defN 1981-01-01 01:01:02 060a09c7 lib/armeabi-v7a/libspake2.so
- -rw-r--r--  0.0 unx  2267028 b-   904101 defN 1981-01-01 01:01:02 86c3f566 lib/x86/libconscrypt_jni.so
+ -rw-r--r--  0.0 unx  2267028 b-   904101 defN 1981-01-01 01:01:02 210cf1bd lib/x86/libconscrypt_jni.so
  -rw-r--r--  0.0 unx    42156 b-    23490 defN 1981-01-01 01:01:02 3ad5af57 lib/x86/libspake2.so
- -rw-r--r--  0.0 unx  2501432 b-   967296 defN 1981-01-01 01:01:02 fa4c4fdf lib/x86_64/libconscrypt_jni.so
+ -rw-r--r--  0.0 unx  2501432 b-   967290 defN 1981-01-01 01:01:02 005f871b lib/x86_64/libconscrypt_jni.so
  -rw-r--r--  0.0 unx    35128 b-    20175 defN 1981-01-01 01:01:02 3259e14c lib/x86_64/libspake2.so

And all the build log says for conscrypt is:

> Task :app:stripReleaseDebugSymbols
Unable to strip the following libraries, packaging them as they are: libconscrypt_jni.so, libspake2.so. Run with --info option to learn more.

I ignored that "stripping error", as resulting file sizes are identical. Funny is, though, that the compressed file sizes slightly differ (as does their hash then of course) – despite both reportedly using the same compression level.

[IzzySoft]

v1.1.0 has the very same issue. I've checked the extracted lib/arm64-v8a/libconscrypt_jni.so: it's not the compression; though both files have identical size, the checksums differ.

v1.0.9 has now been marked "RB failed". We'll have to mark v1.1.0 failed as well, if we cannot get it fixed.

[shiyunjin]

In v1.1.1 I added keepDebugSymbols to prevent stripping libraries, let’s see if this improves the situation.

[IzzySoft]

It does – v1.1.1 was just confirmed to be RB, thanks!

So how do we go from here: keeping them, or trying to find out what goes wrong at stripping?

[shiyunjin]

Just keep it as it is now, conscrypt is an official Google library, binaries are pre-stripped. That's why it prompts unable to strip

https://github.com/google/conscrypt

[IzzySoft]

Ah, OK then. The alternative would have been to build that from source as well 😉