RFR: 7902847: Class directory of a test case should be always used to compile a library [v5]

Leonid Mesnik lmesnik at openjdk.org
Mon Apr 14 17:53:09 UTC 2025


> The classes from test libraries can be compiled implicitly as a test dependency or explicitly with @build tag.
> 
> For first case
> Test library used as a source path during test compilation  and library classes as well as test classes are compiled into class directory. 
> 
> For 2nd case 
> The library classes are compiled using @build tag and library classes are placed into some shared location.
> 
> These 2 cases might be mixed in the single test and can have part of library classes compiled into shared directory and part compiled into test classes directory.
> 
> jtreg uses classfiles to check if source code should be compiled. Let we have 2 classes LibA and Lib that extends LibA.
>  So if class LibA might be  compiled into test class directory while LibB is compiled into library directory. Later  jtreg can try to use LibB class without and compilation because LibB already exists. Thus the CNFE will be thrown.
> 
> The another possible problem, is that the library code might include test code for libraries like "/" and "/vmTestbase"
> so any test classes with same packages would be compiled into the same location because they are treated as library classes. So for tree like this:
> test1/Test.java
> test2/Test.java
> with both tests having
> 
> @library "/"
> @build Test
> 
> we are going just to have one Test class in library directory.
> 
> The only reliable fix would be don't use shared class directory at all and compile all libraries for each test. Although it looks like huge performance overhead, the impact is low. The libraries are not often compiled using build tag. 
> 
> Times for tier1 execution with fix:
> test-results/jtreg_test_lib_test_tier1/text/timeStats.txt:Total elapsed time 0m 28s
> test-results/jtreg_test_hotspot_jtreg_tier1/text/timeStats.txt:Total elapsed time 18m 14s
> test-results/jtreg_test_jdk_tier1/text/timeStats.txt:Total elapsed time 15m 8s
> test-results/jtreg_test_langtools_tier1/text/timeStats.txt:Total elapsed time 7m 59s
> 
> and before fix
> test-results/jtreg_test_lib_test_tier1/text/timeStats.txt:Total elapsed time 0m 32s
> test-results/jtreg_test_hotspot_jtreg_tier1/text/timeStats.txt:Total elapsed time 17m 51s
> test-results/jtreg_test_jdk_tier1/text/timeStats.txt:Total elapsed time 14m 49s
> test-results/jtreg_test_langtools_tier1/text/timeStats.txt:Total elapsed time 7m 56s
> 
> The full fix might require more testing and adding testcase.
> Please note that there are plans to work on the 
> https://bugs.openjdk.org/browse/CODETOOLS-7903882
> and
> https://bugs.openjdk.org/browse/JDK-8346058
> So ...

Leonid Mesnik has updated the pull request incrementally with two additional commits since the last revision:

 - the propery was added to revert sharing of libraries
 - typo fxied

-------------

Changes:
  - all: https://git.openjdk.org/jtreg/pull/256/files
  - new: https://git.openjdk.org/jtreg/pull/256/files/532981df..42b188e6

Webrevs:
 - full: https://webrevs.openjdk.org/?repo=jtreg&pr=256&range=04
 - incr: https://webrevs.openjdk.org/?repo=jtreg&pr=256&range=03-04

  Stats: 36 lines in 4 files changed: 29 ins; 0 del; 7 mod
  Patch: https://git.openjdk.org/jtreg/pull/256.diff
  Fetch: git fetch https://git.openjdk.org/jtreg.git pull/256/head:pull/256

PR: https://git.openjdk.org/jtreg/pull/256


More information about the jtreg-dev mailing list