flan/lib/session.ml
Joseph Ferano a420bb1b1d The session: a program as a live thing
lib/session.ml holds the declarations a running process was built from plus
every change accepted since, which is what an editor needs and what a one-shot
compiler cannot have.

Transactionality came for free. Check.program builds a fresh environment from a
declaration list on every call, so a form that fails to check mutates nothing
and the accumulated list is simply not replaced - no scratch-environment
machinery, which is what I was about to build. Re-checking the whole program
each evaluation costs the frontend, under 10ms, less than the llc after it.
There is a test for the case that matters: a typo, then a good form, in the
same session.

Which names the process was built with comes from the checked program, not from
any accumulated AST, because Check.program prepends the prelude and no AST
contains it. Derive it from declarations and print-line reads as new, gets a
registry cell nobody publishes, and the first call jumps to null.

Three changes are refused with a reason rather than loaded. A function's
signature, because a cell is a bare ptr and every call site compiled before the
change still passes the old arguments through it. A global's type, because the
storage exists and has a shape - reusing it reads at the wrong offsets, and
replacing it discards the state the reload exists to preserve. A struct's
fields, because the values the process is holding have the old layout. Note
what the checker already catches on its own: change a parameter type and the
caller fails to type check first, loudly. These rules only get a turn on a
change the checker accepts, which is a name nothing else in the program uses -
exactly where the silent version lives. Hence an unused defvar and a C-called
defn in the fixtures.

The accumulated list is the post-Load one, so an evaluated import is spliced as
its expansion. Otherwise re-evaluating a file that imports something appends a
second import, Load expands it again, and the duplicate-name pass rejects it.
C-c C-k on sand.flan's own text is the test.

flan reload now takes a program and a file of changed forms rather than a list
of function names and a --new list: the session works out which names are new,
which is the thing a bare CLI could not.

Also fixed, found by running the agent test under load: the agent took SIGPIPE
when a sender read part of a reply and closed. Replies go out with
MSG_NOSIGNAL, per call rather than by installing a handler, because the signal
disposition belongs to the program the agent is embedded in.
2026-09-10 21:48:45 +07:00

197 lines
8.3 KiB
OCaml

(** A live program: the declarations a running process was built from, plus
every change accepted since.
This is what makes an editor possible. [Check.program] builds a fresh
environment from a declaration list on every call, which is exactly the
property a session needs and the reason there is no scratch-environment
machinery here: a form that fails to check leaves nothing behind, because
nothing was mutated. The list is only replaced once the check has
succeeded. Re-checking the whole program each time costs the whole frontend,
which is under 10ms — less than the [llc] that follows it.
Two things the session knows that no single evaluation could:
- **which names the running process was built with.** A name it has is a
symbol the loaded module binds to; a name it lacks goes through the
by-name registry in runtime/flan_dev.c. Getting this wrong is silent:
treating [print-line] as new gives it a registry cell nobody publishes,
and the first call jumps to null. It has to come from the *checked*
program, because [Check.program] prepends the prelude and no accumulated
AST contains it.
- **what the memory of that process looks like.** A cell is a bare pointer
and carries no signature, so a redefined function whose parameters
changed is called by every existing call site with the old ones — no link
error, no trap, a wrong number. Struct fields and global types are the
same class. Those are refused here, with the reason, rather than loaded.
Not here, and deliberately: evaluating an expression. That is a separate
primitive — synthesize a function around the form, call it, render the
value — and it is not what redefining a name is. *)
type t = {
file : string; (* resolves an import's relative path *)
mutable decls : Ast.decl list; (* post-Load: flat, one namespace *)
mutable program : Tast.program; (* the last thing that checked *)
host : Tast.program; (* what the process was built from *)
}
let fail = Loc.fail
let create ~file =
let l = Load.program ~file (Parse.program (Reader.read_file file)) in
let p = Check.program l.Load.decls in
({ file; decls = l.Load.decls; program = p; host = p }, l)
(* A name the running process exports. Everything else is looked up by name at
install time — see [Emit.redefinition]'s [known]. *)
let known t n =
List.exists (fun (f : Tast.fn) -> String.equal f.Tast.name n) t.host.Tast.fns
|| List.exists
(fun (g : Tast.global) -> String.equal g.Tast.gname n)
t.host.Tast.globals
(* ── What a running process cannot be told ─────────────────────────── *)
(* Everything here is a change that would load cleanly and then be wrong. The
house rule (NEXT.md, Watch for) says recognise it and refuse with the
reason, so each one names what it would have broken. *)
let compatible ~loc (old_ : Tast.program) (new_ : Tast.program) =
let find_fn p n =
List.find_opt (fun (f : Tast.fn) -> String.equal f.Tast.name n) p.Tast.fns
in
List.iter
(fun (f : Tast.fn) ->
match find_fn old_ f.Tast.name with
| None -> ()
| Some g ->
let same =
List.length f.Tast.params = List.length g.Tast.params
&& List.for_all2 Types.equal f.Tast.params g.Tast.params
&& Types.equal f.Tast.ret g.Tast.ret
in
(* A cell holds a bare pointer. Every call site compiled before this
change still passes the old arguments through it. *)
if not same then
fail loc
"%s changes signature, from (Fn [%s] %s) to (Fn [%s] %s); \
the calls already compiled into the running program pass the old \
one. Restart to change it."
f.Tast.name
(String.concat " " (List.map Types.to_string g.Tast.params))
(Types.to_string g.Tast.ret)
(String.concat " " (List.map Types.to_string f.Tast.params))
(Types.to_string f.Tast.ret))
new_.Tast.fns;
List.iter
(fun (g : Tast.global) ->
match
List.find_opt
(fun (h : Tast.global) -> String.equal h.Tast.gname g.Tast.gname)
old_.Tast.globals
with
| Some h when not (Types.equal g.Tast.gty h.Tast.gty) ->
(* The storage exists and has a shape. Reusing it for another one
reads fields at the wrong offsets; allocating fresh storage would
silently discard the state the reload exists to preserve. *)
fail loc
"%s changes type, from %s to %s; the running program already laid \
that storage out. Restart to change it."
g.Tast.gname (Types.to_string h.Tast.gty) (Types.to_string g.Tast.gty)
| _ -> ())
new_.Tast.globals;
List.iter
(fun (s : Tast.structure) ->
match
List.find_opt
(fun (r : Tast.structure) -> String.equal r.Tast.sname s.Tast.sname)
old_.Tast.structs
with
| Some r ->
let fields (x : Tast.structure) =
List.map (fun (f : Tast.field) -> (f.Tast.fname, f.Tast.fty))
x.Tast.fields
in
let same =
List.length s.Tast.fields = List.length r.Tast.fields
&& List.for_all2
(fun (an, at) (bn, bt) -> String.equal an bn && Types.equal at bt)
(fields s) (fields r)
in
(* Every value of this type in the running program has the old layout,
including ones held in globals that the reload is preserving. *)
if not same then
fail loc
"%s changes layout; the values the running program is holding have \
the old one. Restart to change it."
s.Tast.sname
| None -> ())
new_.Tast.structs
(* ── Accepting a change ────────────────────────────────────────────── *)
(* The redefinition unit is a list of top-level forms, so this is one path for
both editor commands: C-c C-c sends one form, C-c C-k sends a file. *)
type change = {
ir : string; (* the module to build and send *)
names : string list; (* everything the forms declared *)
fns : string list; (* the subset that has a body to install *)
}
let eval ?(origin = "<eval>") t src : change =
let forms = Reader.read_all ~file:origin src in
(* Through [Load] like any other source, so an evaluated (import ...) means
what it means in a file. Its expansion is what gets spliced, which is also
why the accumulated list is the post-Load one: re-evaluating a file that
imports something would otherwise append a second copy of the import and
the duplicate-name pass would reject it. *)
let incoming = (Load.program ~file:t.file (Parse.program forms)).Load.decls in
let loc =
match incoming with d :: _ -> d.Ast.dloc | [] -> Loc.unknown
in
let names = List.filter_map Ast.declared_name incoming in
let replacement n =
List.find_opt
(fun (d : Ast.decl) -> Ast.declared_name d = Some n)
incoming
in
(* Replaced in place and appended only when genuinely new, so declaration
order — which is emission order for globals — does not shuffle on every
evaluation. *)
let replaced = ref [] in
let kept =
List.map
(fun (d : Ast.decl) ->
match Ast.declared_name d with
| Some n ->
(match replacement n with
| Some nd -> replaced := n :: !replaced; nd
| None -> d)
| None -> d)
t.decls
in
let added =
List.filter
(fun (d : Ast.decl) ->
match Ast.declared_name d with
| Some n -> not (List.exists (String.equal n) !replaced)
| None -> false)
incoming
in
let decls = kept @ added in
(* Nothing above this line has changed the session. A [Loc.Error] from here
leaves it exactly as it was. *)
let program = Check.program decls in
compatible ~loc t.program program;
let fns =
List.filter
(fun n ->
List.exists
(fun (f : Tast.fn) -> String.equal f.Tast.name n)
program.Tast.fns)
names
in
let ir = Emit.redefinition ~dev:true ~known:(known t) program ~fns in
t.decls <- decls;
t.program <- program;
{ ir; names; fns }