Hi,

We hit a linking error (after upgrading to dub 1.26.0). I thought I would try to use dustmite to create a reduced error test case. One week later it is still running (depth 22). I don't suppose there is anyway of determining when it will finish?

On Thursday, 2 September 2021 at 11:04:12 UTC, JG wrote:

Possibly it's not a compiler error at all but a name-mangling change combined with some stale objects in your build.

On Thursday, 2 September 2021 at 11:04:12 UTC, JG wrote:

One way to get a very rough estimate is to take the square of the current reduction (.reduced directory), and divide it by the square of the original source.

For best performance, make sure that:

1. You are using the latest version from https://github.com/CyberShadow/DustMite
2. DustMite itself is built in release mode
3. You are using the -j switch to enable parallelism
4. You are running the reduction on a RAM disk or equivalent

If you don't want to risk losing progress by putting everything (including the result directory) on a RAM disk, the latest bleeding-edge version on the "next" branch introduces a --temp-dir option.

DustMite reductions can be stopped and resumed by pointing DustMite at the latest .reduced directory from a previous run.

On Thursday, 2 September 2021 at 11:20:18 UTC, Vladimir Panteleev wrote:

I meant the square of the size of the respective directory. (bytes / LOC / SLOC...)

If you are still on the first iteration, you may also try switching to the "careful" strategy.

On Thursday, 2 September 2021 at 11:19:55 UTC, jfondren wrote:

Thanks for the suggestion.

I thought I was doing this. What I did was run:

dub clean --all-packages
dub build --force

Then I am getting twelve linking errors. Of the form:
...... error: undefined reference to .....

On Thursday, 2 September 2021 at 17:56:54 UTC, Vladimir Panteleev wrote:

Thanks for the information.

I stopped dustMite and realized I must have done something wrong since the reduced test case doesn't reproduce the problem.

I would really like to try and produce a reduced test case of this problem. However having spent three or four hours trying to figure out what I am doing wrong I think I will have to stop at this point.

On Friday, 3 September 2021 at 19:56:30 UTC, JG wrote:

I tried again. What am I doing wrong?

cp source ~/tmp/source
cd ~/tmp/source
dub build --config prog1 2>&1 | grep "collect2: error: ld returned 1 exit status"
echo $? #produces 0
find . -name *.o -delete
~/d/DustMite/dustmite -j ./ 'dub build --config prog1 2>&1 | grep "collect2: error: ld returned 1 exit status"'
cd ../source.reduced
dub build --config prog1 2>&1 | grep "collect2: error: ld returned 1 exit status"
echo $? #produces 1

On Saturday, 4 September 2021 at 06:18:52 UTC, JG wrote:

First line should be: cp -R source ~/tmp/source

On Saturday, 4 September 2021 at 06:18:52 UTC, JG wrote:

One potential source of trouble is using Dub. Dub doesn't play well with tools such as DustMite, because it keeps state outside of the current directory. In some cases, it even writes to that state, which works very poorly especially with -j. Dub has a built-in "dustmite" command, though it is a bit limited - you could use dub -v to get the compiler and linker commands that Dub runs, and run them directly from your test command.

Some bugs do not reproduce 100% of the time, and are random in nature, due to depending on memory layout or thread timing. Try running the test command a number of times and see if that affects the result.

Regardless, your workflow looks OK from a high level - the source.reduced directory should indeed contain the contents of a directory for which the test command did return 0, for one reason or another.

On Thursday, 2 September 2021 at 11:04:12 UTC, JG wrote:

Hi,
The Dustmite condition you are using seems very generic, therefore this error message can be triggered by various code constellations. This might be the reasons why Dustmite is running so long.

You might delete all fetched dub packages in your home directory and also delete .dub in your project directory. Hopefully this already solves the issue.

Kind regards
Andre