-
Notifications
You must be signed in to change notification settings - Fork 1.7k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[GR-62070] Increased debug info size in comparison to GraalVM for JDK 23 #10512
Comments
FYI @olpaw |
Could be related to #9802 |
Hi @zakkak, Thank you for reaching out to us about this. ![]() ![]() It seems like running things on GraalVM for JDK 25 is a few seconds faster than on GraalVM for JDK 23. |
That's interesting. cc @Karm |
@selhagani I tried reproducing the issue once more and it seems to persist, except for the debug info size which now (that I tested with a newer build) seems to be larger in GraalVM for JDK 23 than in GraalVM for JDK 25. Can you please also check the debug info size of the generated binaries in your case? Apples to Oranges, or GraalVM CE to Oracle GraalVM, comparison.I am getting Java version: 23.0.2+7, vendor version: GraalVM CE 23.0.2+7.1 (
|
I just retested using the latest version of GraalVM for JDK 23 (23.0.2) and the latest available Dev build for JDK 25 and I got the following results: with GraalVM for JDK 23 Community 23.0.2 I get:
with GraalVM CE 25.0.0-dev-20250208_1102 I get:
I will be sharing these results with our team and I'll keep you updated with any news. |
@zakkak in #10512 (comment) you are comparing apples to oranges. To get any meaningful results for comparison for debuginfo generation time and size you have to test Comparing Please provide data for e.g. I do not expect the changes @dominikmascherbauer implemented (that show up in 25) to cause any debuginfo generation performance regressions. Regarding debuginfo size, improvements (smaller size) are expected because he switched output to generate DWARF5 (which is more compact) |
Good catch @olpaw I got carried away with the easy of getting an early build from sdkman that didn't notice that it was Oracle GraalVM instead of CE (and yes I didn't even notice when copy pasting the version over...). Thanks for pointing this out. So back to apples to apples (or as close as I can get to that :) ) Java version: 23.0.2+7, vendor version: GraalVM CE 23.0.2+7.1 (sdk install java 23.0.2-graalce)
Java version: 25+9, vendor version: GraalVM CE 25-dev+9.1 (installed using GraalVM for JDK 25.0.0-dev-20250219_0536 dev build)
The key points remain present:
I wonder if the more compact representation of DWARF5 is the one responsible for the additional overhead. But what I am really curious about is why/how @selhagani seems to be getting exactly the oposite results (in terms of performance not size). |
Different versions of GDB maybe? Before @dominikmascherbauer merged his PR I requested that he checks for such regressions and he also did not see GDB load times to get worse with DWARF5. But even if load times would indeed slightly regress, having smaller |
That's some interesting observations, I did not see such results in my checks. Overall it shows similar results to @zakkak (on my amd64 machine), with slower load times for the updated DWARF5 debug info. GDB 10.2
GDB 11.2
GDB 12.1
GDB 13.2
GDB 14.2
GDB 15.2
GDB 15.2 with
|
In other words, from GDB 13 onwards loading times (for GDB command |
Describe the issue
The debug info generated with the latest graal
master
is larger than that generated with GraalVM for JDK 23, resulting in GDB taking more time to load symbols.Steps to reproduce the issue
Describe GraalVM and your environment:
More details
Locally it takes 27.798 seconds for the 10 iterations when using GraalVM for JDK 23 and 37.218 seconds when using GraalVM for JDK 25.
The text was updated successfully, but these errors were encountered: