Bug 253913

On the GTK and WPE test bots we run with coredumps enabled so when layout tests crash they generate a coredump and we get a backtrace.

The problem is on the EWS bots that we repeat each failure up to 11 times each failure 10 times then a lot of coredumps can be generated in a sort space of time. This may end causing the bot to run out of space. In bug 252762 I implemented a way of cleaning the coredump after processing it but it seems this is not enough as sometimes the bots are still filing the disk.

See: https://ews-build.webkit.org/#/builders/86/builds/5299

The cause is that sometimes crashes happen that are not intercepted by the tooling so those coredump don't get deleted after generating them.

It would be good to investigate why those crashes are not intercepted by the tooling, and is on my todo.

However, at any rate, it makes little sense that we run with coredumps enabled the repeat steps for the EWS. If a test is crashing 11 times we don't need to generate 11 coredumps. It is enough with just picking one.

So I think we should disable coredumps when doing the repeat steps. For doing this we can implement a new flag that can be passed to run-webkit-tests to enable coredumps and we pass this flag on the bots for the steps we want coredumps to be generated.

This can be useful also for developers as they can just enable coredumps when running tests by passing a flag to run-webkit-tests instead of enabling those on the parent sell.

This also has the advantage of having faster runs when doing the retry step with lot of crashes. Having to generate 11 coredumps per crashing test and running GDB on all of them to generate the backtrace is really slow and can make the retry test step last 10x or 20x more time. Generating each backtrace can take easily 1 or 2 minutes so we are spending 10 or 20 minutes per each test that crashes. Not good.