Aggressive tree shaker helps to reduce application footprint by removing all methods that are not referenced in code. This often breaks reflection-based code and required classes has to be explicitly specified using <forceLinkClasses> but sometimes it is not enough. While working with nitrite-database I faced case when class is referenced but some methods of it is still removed. Here is an example:

private void testSerialize() {
        try {
            new ObjectOutputStream(new ByteArrayOutputStream()).writeObject(new HashMap<>());
        } catch (IOException e) {
            e.printStackTrace();
        }
    }

fails with:

java.io.InvalidClassException: java.util.HashMap doesn't have a field loadFactor of type float
	at java.io.ObjectOutputStream.writeFieldValues(ObjectOutputStream.java:955)

This happens due internal private and unreferenced methods got removed from HashMap.

java.io.Serializable contract

As specified in documentation there is special case:

This post continues series of investigations #1 #2 #3. It delivers fix described in previous post: stop/start world changed in way to not loos mach port reference to thread between stop and start world calls. For fix I’ve used old code boehm-gc that is used in MobiVM and not recent one I’ve updated recently just to minimize amount of new problem introduced. I’ll will be back to updated code in one of next snapshots once iOS12 issue is confirmed.

I’ve created a PR1 where all changes can be evaluated.
Pre-built library for testing is also available lib_fix_candidate/libgc.a

Please test and report any issues to gitter channel

This post continues series of investigations #1 #2). Root case – boehm-gc doesn’t resume all threads it paused during GC_stop_world. Details bellow.

UPDATE: there is a follow up on this topic.

Few days ago I narrowed issue to simple case (check this post) but it didn’t answer why this was happening. Today I had time to digg this rock and find pice of code that cause issue. What has been done:

UPDATE: there is a follow up on this topic.
UPDATE2: there is a follow up on this topic.

libgdx developers got concerned about crashes in ios12 beta and there is an issue about it. I was not able to reproduce it using steps provided by Eric Nondahl as was trying to reproduce it on empty iOS system, and following actions helped much:

• switching to other apps will help;
• it is better other apps to be OGL ones;

These steps resulted in reproducing Eric’s POC. But POC) is quite verbose as packed with different cases:

It a follow up of old story. This case hit me back again. This time during a work on keychain utils for linux/windows port. In general this happens in following code:

PKCS12PfxPduBuilder builder = new PKCS12PfxPduBuilder();
...
PKCS12MacCalculatorBuilder macBuilder = new JcePKCS12MacCalculatorBuilder();
PKCS12PfxPdu pfx = builder.build(macBuilder, "".toCharArray());

and the reason in following:

Just have migrated from Mac Mini to MacPro (4 core xeon model mid 2012) to double number of horses. Just swapped SSD from MacMini into MacPro and it is up and running. Build is almost twice faster now, everything seems to be ok but IntelliJ Idea started working extremely slow. Reinstall OS from scratch but it is pure 1 day wasting. It seemed to me that there is no acceleration enabled and reason is not known for me. Probably switch from Intel GPU to ATI might affect existing system setup. Did find that apple.awt.graphics.UseQuartz=true system properties shall enable Quartz acceleration in Java as it is off by default (?). Quick test in Idea and it solves everything !
To add it:

• open Help->Edit Custom VM Options
• add -Dapple.awt.graphics.UseQuartz=true line there
  my working custom options looks like bellow now:

-Xms128m
-Xmx1750m
-XX:ErrorFile=$USER_HOME/java_error_in_idea_%p.log
-XX:HeapDumpPath=$USER_HOME/java_error_in_idea.hprof
-Dapple.awt.graphics.UseQuartz=true

Keep working on Keychain utility for linux-windows port, generating the code signing request for particular. And was surprised that openssl was not able to load it CertificateSigningRequest.certSigningRequest built by Apple:

$ openssl req -text -noout -verify -in CertificateSigningRequest.certSigningRequest
unable to load X509 request
140735602271176:error:0D07209B:asn1 encoding routines:ASN1_get_object:too long:/BuildRoot/Library/Caches/com.apple.xbs/Sources/libressl/libressl-22.50.2/libressl/crypto/asn1/asn1_lib.c:143:
... other complains ...
140735602271176:error:0906700D:PEM routines:PEM_ASN1_read_bio:ASN1 lib:/BuildRoot/Library/Caches/com.apple.xbs/Sources/libressl/libressl-22.50.2/libressl/crypto/pem/pem_oth.c:84:

Apple uses different format ? No, it just writes broken ASN.1 stream. Details bellow:

This post continues series of debugger rework, refer to previous post. In general rework is almost complete and changes are being tested now.
Beside variable resolution problem debug of kotlin code suffered from another issue: debugger was jumping random lines when step over collection iteration with lambdas. Why this was happening is bellow:

History Cannot Use Step Filters while Debugging in Eclipse #309
Fix PR315

Eclipse plugin is on low priority and it debugger there was working on simple cases year ago I was checking. Simple debug session discovered the bug in Event request validation.

Root case

Problem case there was validation of EXCEPTION_ONLY.referenceTypeID event modifier against list of known to debugger ones, but there is case when it can be zero: