Profile: Fix tests when run with nthreads > 1 (for buildkite) (#42270)

Author: IanButterworth

src/signal-handling.c

@@ -37,8 +37,14 @@ void jl_shuffle_int_array_inplace(volatile uint64_t *carray, size_t size, uint64
 
 JL_DLLEXPORT int jl_profile_is_buffer_full(void)
 {
+    // declare buffer full if there isn't enough room to take samples across all threads
+    #if defined(_OS_WINDOWS_)
+        uint64_t nthreads = 1; // windows only profiles the main thread
+    #else
+        uint64_t nthreads = jl_n_threads;
+    #endif
     // the `+ 6` is for the two block terminators `0` plus 4 metadata entries
-    return bt_size_cur + (JL_BT_MAX_ENTRY_SIZE + 1) + 6 > bt_size_max;
+    return bt_size_cur + (((JL_BT_MAX_ENTRY_SIZE + 1) + 6) * nthreads) > bt_size_max;
 }
 
 static uint64_t jl_last_sigint_trigger = 0;

stdlib/Profile/src/Profile.jl

@@ -55,7 +55,8 @@ function init(; n::Union{Nothing,Integer} = nothing, delay::Union{Nothing,Real}
     n_cur = ccall(:jl_profile_maxlen_data, Csize_t, ())
     delay_cur = ccall(:jl_profile_delay_nsec, UInt64, ())/10^9
     if n === nothing && delay === nothing
-        return Int(n_cur), delay_cur
+        nthreads = Sys.iswindows() ? 1 : Threads.nthreads() # windows only profiles the main thread
+        return round(Int, n_cur / nthreads), delay_cur
     end
     nnew = (n === nothing) ? n_cur : n
     delaynew = (delay === nothing) ? delay_cur : delay

stdlib/Profile/test/runtests.jl

@@ -106,15 +106,27 @@ end
 
 @testset "setting sample count and delay in init" begin
     n_, delay_ = Profile.init()
+    n_original = n_
+    nthreads = Sys.iswindows() ? 1 : Threads.nthreads()
+    sample_size_bytes = sizeof(Ptr)
     def_n = Sys.iswindows() && Sys.WORD_SIZE == 32 ? 1_000_000 : 10_000_000
-    @test n_ == def_n
+    if Sys.WORD_SIZE == 32 && (def_n * nthreads * sample_size_bytes) > 2^29
+        @test n_ * nthreads * sample_size_bytes <= 2^29
+    else
+        @test n_ == def_n
+    end
+
     def_delay = Sys.iswindows() && Sys.WORD_SIZE == 32 ? 0.01 : 0.001
     @test delay_ == def_delay
     Profile.init(n=1_000_001, delay=0.0005)
     n_, delay_ = Profile.init()
-    @test n_ == 1_000_001
+    if Sys.WORD_SIZE == 32 && (1_000_001 * nthreads * sample_size_bytes) > 2^29
+        @test n_ * nthreads * sample_size_bytes <= 2^29
+    else
+        @test n_ == 1_000_001
+    end
     @test delay_ == 0.0005
-    Profile.init(n=def_n, delay=def_delay)
+    Profile.init(n=n_original, delay=def_delay)
 end
 
 @testset "warning for buffer full" begin