Develop the film last, in the view transform's place

A film stock ran at order 25, after white balance and exposure, and
everything after it — contrast, the curves, the colour mixer, the
grading, every mask layer's tone — acted on the film's display-referred
output, as though the frame had been scanned and then worked on. That
was a display-referred rendering in the middle of the chain, which D19
removes (FR-DEV-3f, FR-DEV-3j).

`film_sim` is now in `Stage::View` beside `view_transform`. While a
stock is loaded the composer emits it in the view transform's place and
not the sigmoid; otherwise the sigmoid. So every edit is a decision
about the exposure the negative receives, and the film is the last
thing that happens to the picture — after the detail stage too, in the
view pass, which already binds the film's tables, the mask array and
the grain's source position. Its per-layer settings blend there as they
did in the fused pass.

`op_renders` goes: a rendering is chosen, not suppressed, and the only
render with no view transform is the camera-space tap. The YAML order
moves to 190 so the panel reads in pipeline order; the stage, not the
number, is what places it.

Existing edits that combine a stock with tone or colour operations now
render differently: those operations used to act on the print, and now
act on the scene.

core/dr-gpu/src/adjust.rs

@@ -2469,7 +2469,14 @@ mod tests {
                 assert!(!shader.source.contains("---- view_transform ----"));
                 continue;
             }
-            let point = shader.source.contains(&format!("---- {id} ----"));
+            // A view operation is in the view pass when a detail stage
+            // follows, which it does here (D19).
+            let block = format!("---- {id} ----");
+            let point = shader.source.contains(&block)
+                || shader
+                    .view
+                    .as_ref()
+                    .is_some_and(|v| v.source.contains(&block));
             let neighbourhood = detail
                 .passes
                 .iter()
@@ -2503,8 +2510,15 @@ mod tests {
         // because the two catch different faults: the XOR catches an operation
         // in the wrong stage, this catches a block in the shader that nothing
         // in the chain asked for.
+        //
+        // The view pass repeats the prologue — framing and the warps — for
+        // the positions it publishes, so only its operation blocks count.
+        let view_blocks = shader
+            .view
+            .as_ref()
+            .map_or(0, |v| v.source.matches("---- ").count() - (warp_blocks + 1));
         assert_eq!(
-            shader.source.matches("---- ").count(),
+            shader.source.matches("---- ").count() + view_blocks,
             fused_blocks + warp_blocks + 1,
             "the fused shader carries a block nothing in the chain asked for"
         );

core/dr-pipeline/ops/README.md

@@ -240,17 +240,16 @@ coefficients that are not parameters.
 `Operation::renders`, and it is worth knowing why before writing a second one.
 
 Every other node *adjusts* a picture. That one *makes* it: a film stock's
-characteristic curve does the camera profile's base curve's job, from
-measurements rather than from a curve somebody drew. Running both renders the
-scene twice — the camera's rendering, and then a film's rendering of *that* —
-which looks like neither and reads as a colour-management bug with no
-colour-management bug to find.
+characteristic curve does the view transform's job, from measurements rather
+than from a curve somebody chose. Running both renders the scene twice — the
+default rendering, and then a film's rendering of *that* — which looks like
+neither and reads as a colour-management bug with no colour-management bug to
+find.
 
-So a node declaring `renders` hands back display-referred linear sRGB, and in
-exchange the composer does not emit the base curve. It is handed working-space
-colour like every other node: the conversion out of camera space is no longer
-the rendering's to take over, because since D19 it runs before every node but
-white balance (see "Stages" below). `distortion` and
+So `film_sim` is in `Stage::View` beside `view_transform`, and while a stock
+is loaded the composer emits it in the view transform's place, last, after the
+detail stage, and not the sigmoid (D19). It is handed working-space colour and
+hands back display-referred linear sRGB for the output transform. `distortion` and
 `aberration` are `Warp`s rather than operations: they rewrite coordinates
 before sampling rather than transforming a colour after it.
 

core/dr-pipeline/ops/film_sim.yaml

@@ -1,5 +1,5 @@
 id: film_sim
-order: 25
+order: 190
 # What this node is *about* is not written here, and cannot be: a `rust:` node
 # publishes its own descriptor, so `attributes:` in this file would be read,
 # validated and then ignored. See `Attribute::Effect` on `FilmSim`'s descriptor
@@ -15,11 +15,13 @@ why_rust: |
   and the conversion out of camera space on its behalf.
 
 placement: |
-  After white balance and exposure, and before everything else.
+  Last, in the view transform's place (D19, FR-DEV-3j), after every other
+  operation and after the detail stage.
 
-  Those two are what the camera did — interpreting the sensor, and correcting
-  the amount of light that reached it — and they are only meaningful on
-  scene-linear values, which is what a film has to be handed. Everything below
-  is a decision about the picture, and a decision about the picture belongs
-  after the film has rendered it, exactly as it does when you scan a frame and
-  then work on the scan.
+  Before D19 it sat at order 25, after white balance and exposure, and every
+  decision below it acted on the film's output, as though the frame had been
+  scanned and then worked on. That put a display-referred rendering in the
+  middle of the chain, which is what D19 removes: every operation is now handed
+  the scene, and the film is the last thing that happens to the picture — an
+  edit is a decision about the exposure the negative receives. `Stage::View`
+  is what puts it there; this number only places it in the panel's order.

core/dr-pipeline/src/lib.rs

@@ -176,7 +176,14 @@ mod tests {
                 assert!(!shader.source.contains("---- view_transform ----"));
                 continue;
             }
-            let point = shader.source.contains(&format!("---- {id} ----"));
+            // A view operation is in the view pass when a detail stage
+            // follows, which it does here (D19).
+            let block = format!("---- {id} ----");
+            let point = shader.source.contains(&block)
+                || shader
+                    .view
+                    .as_ref()
+                    .is_some_and(|v| v.source.contains(&block));
             let neighbourhood = detail
                 .passes
                 .iter()
@@ -189,8 +196,12 @@ mod tests {
             );
             fused_blocks += usize::from(point);
         }
+        let view_blocks = shader
+            .view
+            .as_ref()
+            .map_or(0, |v| v.source.matches("---- ").count());
         assert_eq!(
-            shader.source.matches("---- ").count(),
+            shader.source.matches("---- ").count() + view_blocks,
             fused_blocks,
             "the fused shader carries a block nothing in the chain asked for"
         );

core/dr-pipeline/src/operation.rs

@@ -337,18 +337,17 @@ pub trait Operation: Send + Sync {
     /// Whether this operation *is* the rendering, rather than an adjustment to
     /// one.
     ///
-    /// Almost everything returns `false`. An operation that returns `true`
-    /// hands back display-referred linear sRGB, and in exchange the composer
-    /// does not emit the camera profile's base curve — because this operation
-    /// has done its job. It is handed working-space colour like any other
-    /// scene-stage operation (D19).
+    /// Almost everything returns `false`. A [`Stage::View`] operation that
+    /// returns `true` is a complete rendering: while it is active the composer
+    /// emits it in the view transform's place and not the default view
+    /// transform beside it (FR-DEV-3j).
     ///
     /// The reason it is a trait method and not a flag the caller sets is the
     /// one [`compose_full`] gives for deciding the output mode the same way: a
     /// caller that got it wrong would produce a shader that compiles, runs, and
     /// renders the picture twice. `film_sim` is the operation this exists for —
-    /// a stock's characteristic curve does the base curve's job, from
-    /// measurements, and running both is the camera's rendering of the scene
+    /// a stock's characteristic curve does the view transform's job, from
+    /// measurements, and running both is the default rendering of the scene
     /// followed by a film's rendering of *that*.
     fn renders(&self) -> bool {
         false
@@ -469,7 +468,10 @@ pub enum Stage {
     /// unbounded. Nothing here clamps above 1.0 or encodes (ARCH §6.14).
     Scene,
     /// The view transform (FR-DEV-3j): after every scene operation, and the
-    /// one stage allowed to map the scene to a display range.
+    /// one stage allowed to map the scene to a display range. `view_transform`
+    /// and `film_sim` are its two operations, and one of them runs: the film
+    /// while a stock is loaded (see [`Operation::renders`]), the sigmoid
+    /// otherwise.
     ///
     /// Composed **whatever its state**. A neutral operation is otherwise left
     /// out, but a photograph with no view transform is a scan rather than a
@@ -849,11 +851,6 @@ fn compose_inner(
         )
     };
 
-    // Whether an operation has taken over the rendering. Decided from the
-    // operations for the same reason `output_mode` is: a caller that got it
-    // wrong would produce a shader that compiles and renders the picture twice.
-    let op_renders = ops.iter().any(|o| o.is_active() && o.renders());
-
     let mut uniform_fields = String::new();
     let mut uniform_values: Vec<f32> = Vec::new();
     let mut body = String::new();
@@ -882,8 +879,8 @@ fn compose_inner(
 
     // TRACES: FR-DEV-3j
     // Whether the composer emits a view transform — which is every render but
-    // the camera-space tap and one a rendering operation has taken over.
-    let views = !op_renders && output_mode != OutputMode::CameraLinear;
+    // the camera-space tap, whose job is the sensor's own numbers.
+    let views = output_mode != OutputMode::CameraLinear;
     // And whether *this* shader is where it goes. Not the fused pass when a
     // detail stage follows: the view transform maps into a display range, and
     // a sharpener handed a display range is what D19 set out to stop. The
@@ -940,13 +937,18 @@ fn compose_inner(
             .map(|o| o.as_ref())
             .filter(move |o| o.detail().is_none() && o.stage() == stage)
     };
-    // The view operation, or a default one when the caller's chain holds
-    // none — `compose(&[])` in a test, or a probe. Absent altogether where
-    // nothing is to be rendered: see `views`.
+    // The view operation: an active rendering — a loaded film stock — if
+    // there is one, otherwise the chain's view transform, or a default one
+    // when the caller's chain holds none (`compose(&[])` in a test, or a
+    // probe). Decided from the operations for the same reason `output_mode`
+    // is: a caller that got it wrong would produce a shader that compiles and
+    // renders the picture twice. Absent altogether where nothing is to be
+    // rendered: see `views`.
     let default_view = crate::ops::ViewTransform::new();
     let view: Option<&dyn Operation> = views_here.then(|| {
         point(Stage::View)
-            .next()
+            .find(|o| o.is_active() && o.renders())
+            .or_else(|| point(Stage::View).find(|o| !o.renders()))
             .unwrap_or(&default_view as &dyn Operation)
     });
 
@@ -1190,7 +1192,7 @@ fn compose_inner(
     let rendering_tail = if views {
         String::new()
     } else {
-        "    // No view transform: an operation declaring `Operation::renders` has\n    // mapped the scene to a display range itself, or this is the\n    // camera-space tap, which stores the sensor's own numbers.\n"
+        "    // No view transform: this is the camera-space tap, which stores the\n    // sensor's own numbers.\n"
             .to_string()
     };
 
@@ -2319,10 +2321,22 @@ mod tests {
         }));
         assert!(film.is_active(), "the fixture did not load");
 
-        let source = compose(&[Box::new(film) as Box<dyn Operation>]).source;
+        // Listed first, and emitted last anyway: a stock is the view
+        // transform (D19), so graph order does not put it among the scene
+        // operations the way order 25 once did.
+        let source = compose(&[
+            Box::new(film) as Box<dyn Operation>,
+            fake(DESC_A.clone(), 2.0, false),
+        ])
+        .source;
         let film_at = source
             .find("---- film_sim ----")
             .expect("the operation itself must still be emitted");
+        let scene_op = source.find("---- op_a ----").expect("op present");
+        assert!(
+            scene_op < film_at,
+            "the film must see every scene operation's result"
+        );
         assert!(
             !source.contains("view_sigmoid"),
             "the view transform is still being applied on top of the film"

core/dr-pipeline/src/ops/film_sim.rs

@@ -1,22 +1,25 @@
 //! TRACES: FR-DEV-3f
 //! Film simulation — the stock renders the picture.
 //!
-//! # Why this one replaces the base curve
+//! # Why this one is the view transform
 //!
-//! [`crate::ops`]' other nodes adjust a picture. This one *makes* it. The base
-//! curve exists because sensor data is scene-referred and nothing anybody looks
-//! at is (FR-DEV-3e); a film stock's characteristic curve does the same job,
-//! from measurements, with a toe and a shoulder that were coated onto acetate
-//! rather than drawn. Running both renders the image twice — the camera's
-//! JPEG-ish rendering, and then a film's rendering of that — which is not what
+//! [`crate::ops`]' other nodes adjust a picture. This one *makes* it. The view
+//! transform exists because sensor data is scene-referred and nothing anybody
+//! looks at is (FR-DEV-3j); a film stock's characteristic curve does the same
+//! job, from measurements, with a toe and a shoulder that were coated onto
+//! acetate rather than drawn. Running both renders the image twice — the
+//! default rendering, and then a film's rendering of that — which is not what
 //! either is for and looks like neither.
 //!
-//! So this node declares [`Operation::renders`], and the composer answers by
-//! emitting neither the base curve nor the camera matrix. Both jobs move here:
-//! the fragment takes camera RGB, converts it to linear sRGB itself with the
-//! matrix already in the uniform block, and returns linear sRGB. That is a
-//! contract worth stating plainly, because a node that got half of it wrong
-//! would produce a picture that renders perfectly and is wrong everywhere.
+//! So this node is in [`Stage::View`] and declares [`Operation::renders`]: when
+//! a stock is loaded the composer puts it at the end of the chain in place of
+//! the default sigmoid (D19). It is handed working-space colour — linear sRGB
+//! primaries, scene-referred, after every other operation and after the detail
+//! stage — and returns display-referred linear sRGB for the output transform.
+//! Before D19 it ran at order 25, after exposure and before everything else,
+//! and the operations below it acted on its output. They now act on the scene
+//! it is shown: an edit is a decision about the exposure the negative
+//! receives, and the film is the last thing that happens to the picture.
 //!
 //! # Why the tables are not parameters
 //!
@@ -34,7 +37,7 @@
 use std::sync::{Arc, LazyLock};
 
 use crate::descriptor::{Attribute, LocalizedKey, OpDescriptor, OpId, ParamDescriptor, ParamId};
-use crate::operation::{Operation, Uniform};
+use crate::operation::{Operation, Stage, Uniform};
 
 pub const ID: OpId = OpId("film_sim");
 pub const EXPOSURE: ParamId = ParamId("exposure");
@@ -304,11 +307,17 @@ impl Operation for FilmSim {
         self.tables.is_some()
     }
 
-    /// This node renders; the camera's own rendering must not also run.
+    /// This node renders; the default view transform must not also run.
     fn renders(&self) -> bool {
         true
     }
 
+    /// TRACES: FR-DEV-3f | FR-DEV-3j
+    /// The view transform's place, at the end of the chain (D19).
+    fn stage(&self) -> Stage {
+        Stage::View
+    }
+
     fn set_film_tables(&mut self, tables: Option<&FilmTables>) {
         self.set_tables(tables.cloned());
     }