Avoid using separate tasks for code evaluation

src/QuartoNotebookWorker/src/Malt.jl

@@ -119,31 +119,29 @@ interrupt(::Nothing) = nothing
 function handle(::Val{MsgType.from_host_call_with_response}, socket, msg, msg_id::MsgID)
     f, args, kwargs, respond_with_nothing = msg
 
-    @async begin
-        result, success = try
-            result = f(args...; kwargs...)
-
-            # @debug("WORKER: Evaluated result", result)
-            (respond_with_nothing ? nothing : result, true)
-        catch err
-            # @debug("WORKER: Got exception!", e)
-            (format_error(err, catch_backtrace()), false)
-        end
-
-        _serialize_msg(
-            socket,
-            success ? MsgType.from_worker_call_result : MsgType.from_worker_call_failure,
-            msg_id,
-            result,
-        )
+    result, success = try
+        result = f(args...; kwargs...)
+
+        # @debug("WORKER: Evaluated result", result)
+        (respond_with_nothing ? nothing : result, true)
+    catch err
+        # @debug("WORKER: Got exception!", e)
+        (format_error(err, catch_backtrace()), false)
     end
+
+    _serialize_msg(
+        socket,
+        success ? MsgType.from_worker_call_result : MsgType.from_worker_call_failure,
+        msg_id,
+        result,
+    )
 end
 
 
 function handle(::Val{MsgType.from_host_call_without_response}, socket, msg, msg_id::MsgID)
     f, args, kwargs, _ignored = msg
 
-    @async try
+    try
         f(args...; kwargs...)
     catch e
         @warn(

src/server.jl

@@ -191,14 +191,6 @@ end
 
 # Implementation.
 
-function remote_eval_fetch_channeled(worker, expr)
-    code = quote
-        put!(stable_execution_task_channel_in, $(QuoteNode(expr)))
-        take!(stable_execution_task_channel_out)
-    end
-    return Malt.remote_eval_fetch(worker, code)
-end
-
 function init!(file::File, options::Dict)
     worker = file.worker
     Malt.remote_eval_fetch(worker, worker_init(file, options))
@@ -218,7 +210,7 @@ function refresh!(file::File, options::Dict)
         file.output_chunks = []
         init!(file, options)
     end
-    remote_eval_fetch_channeled(file.worker, :(refresh!($(options)); revise_hook()))
+    Malt.remote_eval_fetch(file.worker, :(refresh!($(options)); revise_hook()))
 end
 
 function _cache_file(f::File, source_code_hash)
@@ -930,7 +922,7 @@ function evaluate_raw_cells!(
                     $(chunk.cell_options),
                 ))
 
-                worker_results, expand_cell = remote_eval_fetch_channeled(f.worker, expr)
+                worker_results, expand_cell = Malt.remote_eval_fetch(f.worker, expr)
 
                 # When the result of the cell evaluation is a cell expansion
                 # then we insert the original cell contents before the expanded
@@ -1141,7 +1133,7 @@ function evaluate_raw_cells!(
                             # inline evaluation since you can't pass cell
                             # options and so `expand` will always be `false`.
                             worker_results, expand_cell =
-                                remote_eval_fetch_channeled(f.worker, expr)
+                                Malt.remote_eval_fetch(f.worker, expr)
                             expand_cell && error("inline code cells cannot be expanded")
                             remote = only(worker_results)
                             if !isnothing(remote.error)
@@ -1202,7 +1194,7 @@ function evaluate_params!(f, params::Dict{String})
         :(@eval getfield(Main, :Notebook) const $(Symbol(key::String)) = $value)
     end
     expr = Expr(:block, exprs...)
-    remote_eval_fetch_channeled(f.worker, expr)
+    Malt.remote_eval_fetch(f.worker, expr)
     return
 end
 

src/startup.jl

@@ -145,21 +145,6 @@ end
 render(args...; kwargs...) = QuartoNotebookWorker.render(args...; kwargs...)
 revise_hook() = @static QUARTO_ENABLE_REVISE ? QuartoNotebookWorker.revise_hook() : nothing
 
-# Issue #192
-#
-# Malt itself uses a new task for each `remote_eval` and because of
-# this, random number streams are not consistent across runs even if
-# seeded, as each task introduces a new state for its task-local RNG.
-# As a workaround, we feed all `remote_eval` requests through these
-# channels, such that the task executing code is always the same.
-const stable_execution_task_channel_out = Channel()
-const stable_execution_task_channel_in = Channel() do chan
-    for expr in chan
-        result = Core.eval(Main, expr)
-        put!(stable_execution_task_channel_out, result)
-    end
-end
-
 # Step 6:
 #
 # Ensures that the LOAD_PATH is returned to it's previous state without the