Restart threads on DAP evaluation

[st0012]

This makes sure DAP evaluation request also doesn't hang.

See #947 for the original console implementation.

[st0012]

@ono-max I tried implementing the same change for CDP as well but it still hangs. So I will leave that to you.

Note: I discovered some issues on a test helper when doing the above and opened #951 for it.

[ko1]

Thank you for the report. But this patch is not acceptable because it doesn't stop other threads after the evaluation.

_th0=Thread.new{loop{sleep0.5;p:th0}}_th1=Thread.new{loop{sleep0.5;p:th1}}loopdosleep0.5p:mainend

Please try the above code and evaluate something on the debug console.

[ko1]

[master]$ git diff
diff --git a/lib/debug/server_dap.rb b/lib/debug/server_dap.rb
index 8298428..bae1e4d 100644
--- a/lib/debug/server_dap.rb+++ b/lib/debug/server_dap.rb@@ -625,6 +625,7 @@ module DEBUGGER__
expr = req.dig('arguments', 'expression')
if find_waiting_tc(tid)
+ restart_all_threads
request_tc [:dap, :evaluate, req, fid, expr, context]
else
fail_response req
@@ -701,6 +702,7 @@ module DEBUGGER__
register_vars result[:variables], tid
@ui.respond req, result
when :evaluate
+ stop_all_threads
message = result.delete :message
if message
@ui.respond req, success: false, message: message

could you add stop_all_threads?

[st0012]

Ah I thought it's handled in Session#process_event too. I've updated code but will try to put up a test case for thread-stopping too.

[st0012]

I think I found another issue: when I run the test, the result was not deterministic even without sleep and fails around 10% of the time on my machine.

I initially thought it's because of my test, but then I found that it's because this early return isn't always correct. Sometimes it doesn't reflect the actual running thread count and thus the debugger doesn't stop the threads. If I comment it out, the test doesn't fail intermittently anymore. Should I address that in this PR too?

[ko1]

Now current behavior is already non-deterministic (by design).
Essentially this is hard to write a test for such cases.

[st0012]

As the test could make CI unstable, I've dropped it.