Poor performance when overrunning a large obfuscation dictionary

[quat1024]

In one project, we use an obfuscation dictionary with 10,000 entries. We found that most CPU time was spent in proguard.obfuscate.DictionaryNameFactory.nextName.

It turns out we had more than 10,000 symbols to obfuscate in the project and ran out the dictionary. DictionaryNameFactory falls back to generating fresh names and checking them against the dictionary so it doesn't generate a duplicate, but this duplicate-checking uses a linear scan (List#contains, line 255), so our large dictionary was causing this code path to be very slow.

proguard/base/src/main/java/proguard/obfuscate/DictionaryNameFactory.java

Lines 250 to 255 in 3a9b11b

	// Return the next different name from the other name factory.
	do
	{
	name = nameFactory.nextName();
	}
	while (names.contains(name));

[phase]

here's a profiler snapshot from VisualVM:

by expanding our dictionary from 10k to 50k, the problem went away and the time taken by proguard cut in half.

after this was solved, another profile shows another hot loop is taking up a significant portion of time

https://github.com/Guardsquare/proguard-core/blob/9ae93fd664f8b92830e9ea402a65c0ffd6f73dce/base/src/main/java/proguard/classfile/util/MethodLinker.java#L150-L158

[mrjameshamilton]

The dictionary performance should be improved in 7.6 with the following commit: 03d7eff