Make sure we don't clean up an active handle during a next download.

[maleadt]

Fixes #99

[codecov-io]

Codecov Report

Merging #101 (b8d1ea0) into master (af6864d) will increase coverage by 0.02%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #101      +/-   ##
==========================================
+ Coverage   92.96%   92.98%   +0.02%     
==========================================
  Files           5        5              
  Lines         483      485       +2     
==========================================
+ Hits          449      451       +2     
  Misses         34       34              

Impacted Files	Coverage Δ
src/Curl/Multi.jl	95.83% <100.00%> (+0.07%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update af6864d...b8d1ea0. Read the comment docs.

[giordano]

Is any of the tests in #99 reproducible in CI?

[maleadt]

Yes, I've added the MWE as a test.

[fredrikekre]

Does it also fix #95 or are they completely separate issues? Have you tried enabling HTTP/2 wich this patch?

[maleadt]

Does it also fix #95 or are they completely separate issues? Have you tried enabling HTTP/2 wich this patch?

That seems very possible, but I couldn't reproduce with HTTP 2.0 re-enabled (and this patch either disabled or enabled). Could you try on your system?

[StefanKarpinski]

Seems better to just hoist the uv_timer_stop(multi.timer) call out of init!.

[StefanKarpinski]

I.e. something like this:

diff --git a/src/Curl/Multi.jl b/src/Curl/Multi.jl
index f16e365..2a5f538 100644
--- a/src/Curl/Multi.jl
+++ b/src/Curl/Multi.jl
@@ -18,7 +18,6 @@ mutable struct Multi
 end

 function init!(multi::Multi)
-    uv_timer_stop(multi.timer)
     multi.handle = curl_multi_init()
     add_callbacks(multi)
     set_defaults(multi)
@@ -43,7 +42,12 @@ end
 function add_handle(multi::Multi, easy::Easy)
     lock(multi.lock) do
         isempty(multi.easies) && preserve_handle(multi)
-        multi.handle == C_NULL && init!(multi)
+        if multi.handle == C_NULL
+            init!(multi)
+        else
+            # stop grace timer
+            uv_timer_stop(multi.timer)
+        end
         push!(multi.easies, easy)
         @check curl_multi_add_handle(multi.handle, easy.handle)
     end

The remaining question is whether the call to uv_timer_stop should happen on every added handle or only the first one (when multi.handle == C_NULL). I suspect the latter, but I'm not 100% sure. If so, then this code was calling uv_timer_stop exactly when it shouldn't and not calling it when it should 😬