Add option to snoop commands to control spawning separate process vs eval'ing in existing process

SnoopCompileCore/src/SnoopCompileCore.jl

@@ -23,4 +23,8 @@ if VERSION >= v"1.6.0-DEV.1192"  # https://github.com/JuliaLang/julia/pull/37136
     include("snoopl.jl")
 end
 
+if VERSION >= v"1.6.0-DEV.1192"  # https://github.com/JuliaLang/julia/pull/37136
+    include("snoop_all.jl")
+end
+
 end

SnoopCompileCore/src/snoop_all.jl

@@ -0,0 +1,11 @@
+# TODO: Add `@snoopc` into the mix
+macro snoop_all(csv_f, yaml_f, commands)
+    esc(quote
+        v = @snoopi_deep begin
+            @snoopl pspawn=false $csv_f $yaml_f begin
+                $commands
+            end
+        end;
+        v
+    end)
+end

SnoopCompileCore/src/snoopl.jl

@@ -4,7 +4,7 @@ using Serialization
 
 """
 ```
-@snoopl "func_names.csv" "llvm_timings.yaml" begin
+@snoopl [jlflags="..."] [pspawn=true] "func_names.csv" "llvm_timings.yaml" begin
     # Commands to execute, in a new process
 end
 ```
@@ -14,16 +14,26 @@ be used for the input to `SnoopCompile.read_snoopl("func_names.csv", "llvm_timin
 
 The logs contain the amount of time spent optimizing each "llvm module", and information
 about each module, where a module is a collection of functions being optimized together.
+
+If `pspawn=false`, the commands will be run in the same julia process, via `eval()` in
+the current module. This will only report LLVM optimizations for _new_ compilations, that
+haven't already been cached in this process, which can be (carefully) used to pruned down
+the results to only the code you are interested in.
 """
-macro snoopl(flags, func_file, llvm_file, commands)
-    return :(snoopl($(esc(flags)), $(esc(func_file)), $(esc(llvm_file)), $(QuoteNode(commands))))
+macro snoopl(args...)
+    @assert length(args) >= 3 """Usage: @snoopl [args...] "snoopl.csv" "snoopl.yaml" commands"""
+    flags, (func_file, llvm_file, commands) = args[1:end-3], args[end-2:end]
+    flags = [esc(e) for e in flags]
+    return :(snoopl($(esc(func_file)), $(esc(llvm_file)), $(QuoteNode(commands)), $__module__; $(flags...)))
 end
 macro snoopl(func_file, llvm_file, commands)
-    return :(snoopl(String[], $(esc(func_file)), $(esc(llvm_file)), $(QuoteNode(commands))))
+    return :(snoopl($(esc(func_file)), $(esc(llvm_file)), $(QuoteNode(commands)), $__module__))
 end
 
-function snoopl(flags, func_file, llvm_file, commands)
-    println("Launching new julia process to run commands...")
+function snoopl(func_file, llvm_file, commands, _module; pspawn=true, jlflags="")
+    if pspawn
+        println("Launching new julia process to run commands...")
+    end
     # addprocs will run the unmodified version of julia, so we
     # launch it as a command.
     code_object = """
@@ -32,23 +42,31 @@ function snoopl(flags, func_file, llvm_file, commands)
                 Core.eval(Main, deserialize(stdin))
             end
             """
-    process = open(`$(Base.julia_cmd()) $flags --eval $code_object`, stdout, write=true)
-    serialize(process, quote
+    record_and_run_quote = quote
         let func_io = open($func_file, "w"), llvm_io = open($llvm_file, "w")
             ccall(:jl_dump_emitted_mi_name, Nothing, (Ptr{Nothing},), func_io.handle)
             ccall(:jl_dump_llvm_opt, Nothing, (Ptr{Nothing},), llvm_io.handle)
             try
-                $commands
+                @eval $commands
             finally
                 ccall(:jl_dump_emitted_mi_name, Nothing, (Ptr{Nothing},), C_NULL)
                 ccall(:jl_dump_llvm_opt, Nothing, (Ptr{Nothing},), C_NULL)
                 close(func_io)
                 close(llvm_io)
             end
         end
-        exit()
-    end)
-    wait(process)
-    println("done.")
+    end
+
+    if pspawn
+        process = open(`$(Base.julia_cmd()) $jlflags --eval $code_object`, stdout, write=true)
+        serialize(process, quote
+            $record_and_run_quote
+            exit()
+        end)
+        wait(process)
+        println("done.")
+    else
+        Core.eval(_module, record_and_run_quote)
+    end
     nothing
 end