Fix for Pixmap.set_pixel().

Also added optimisation in src.extra.i.

Addresses #3050.

Author: julian-smith-artifex-com

src/__init__.py

@@ -9647,6 +9647,11 @@ def __init__(self, *args):
                 text += f'    {type(arg)}: {arg}\n'
             raise Exception( text)
 
+        # 2024-01-16: Experimental support for a memory-view of the underlying
+        # data.  Doesn't seem to make much difference to Pixmap.set_pixel() so
+        # not currently used.
+        self._memory_view = None
+
     def __len__(self):
         return self.size
 
@@ -10117,20 +10122,28 @@ def set_origin(self, x, y):
 
     def set_pixel(self, x, y, color):
         """Set color of pixel (x, y)."""
+        if g_use_extra:
+            return extra.set_pixel(self.this.m_internal, x, y, color)
         pm = self.this
         if not _INRANGE(x, 0, pm.w() - 1) or not _INRANGE(y, 0, pm.h() - 1):
             raise ValueError( MSG_PIXEL_OUTSIDE)
         n = pm.n()
-        c = list()
         for j in range(n):
             i = color[j]
             if not _INRANGE(i, 0, 255):
                 raise ValueError( MSG_BAD_COLOR_SEQ)
-            c.append( ord(i))
         stride = mupdf.fz_pixmap_stride( pm)
         i = stride * y + n * x
-        for j in range(n):
-            pm.m_internal.samples[i + j] = c[j]
+        if 0:
+            # Using a cached self._memory_view doesn't actually make much
+            # difference to speed.
+            if not self._memory_view:
+                self._memory_view = self.samples_mv
+            for j in range(n):
+                self._memory_view[i + j] = color[j]
+        else:
+            for j in range(n):
+                pm.fz_samples_set(i + j, color[j])
 
     def set_rect(self, bbox, color):
         """Set color of all pixels in bbox."""
@@ -10154,6 +10167,8 @@ def shrink(self, factor):
             JM_Warning("ignoring shrink factor < 1")
             return
         mupdf.fz_subsample_pixmap( self.this, factor)
+        # Pixmap has changed so clear our memory view.
+        self._memory_view = None
 
     @property
     def size(self):

src/extra.i

@@ -3670,6 +3670,7 @@ PyObject* extractBLOCKS(mupdf::FzStextPage& self)
             mupdf::fz_clear_buffer(res);  // set text buffer to empty
             int line_n = -1;
             int last_char = 0;
+            (void) line_n;  /* Not actually used, but keeping in the code for now. */
             for (fz_stext_line* line = block->u.t.first_line; line; line = line->next)
             {
                 line_n++;
@@ -3954,6 +3955,62 @@ int pixmap_n(mupdf::FzPixmap& pixmap)
     return mupdf::fz_pixmap_components( pixmap);
 }
 
+static int
+JM_INT_ITEM(PyObject *obj, Py_ssize_t idx, int *result)
+{
+    PyObject *temp = PySequence_ITEM(obj, idx);
+    if (!temp) return 1;
+    if (PyLong_Check(temp)) {
+        *result = (int) PyLong_AsLong(temp);
+        Py_DECREF(temp);
+    } else if (PyFloat_Check(temp)) {
+        *result = (int) PyFloat_AsDouble(temp);
+        Py_DECREF(temp);
+    } else {
+        Py_DECREF(temp);
+        return 1;
+    }
+    if (PyErr_Occurred()) {
+        PyErr_Clear();
+        return 1;
+    }
+    return 0;
+}
+
+PyObject *set_pixel(fz_pixmap* pm, int x, int y, PyObject *color)
+{
+    fz_context* ctx = mupdf::internal_context_get();
+    if (0
+            || x < 0
+            || x >= pm->w
+            || y < 0
+            || y >= pm->h
+            )
+    {
+        throw std::range_error( MSG_PIXEL_OUTSIDE);
+    }
+    int n = pm->n;
+    if (!PySequence_Check(color) || PySequence_Size(color) != n) {
+        throw std::range_error(MSG_BAD_COLOR_SEQ);
+    }
+    int i, j;
+    unsigned char c[5];
+    for (j = 0; j < n; j++) {
+        if (JM_INT_ITEM(color, j, &i) == 1) {
+            throw std::range_error(MSG_BAD_COLOR_SEQ);
+        }
+        if (i < 0 or i >= 256) {
+            throw std::range_error(MSG_BAD_COLOR_SEQ);
+        }
+        c[j] = (unsigned char) i;
+    }
+    int stride = fz_pixmap_stride(ctx, pm);
+    i = stride * y + n * x;
+    for (j = 0; j < n; j++) {
+        pm->samples[i + j] = c[j];
+    }
+    Py_RETURN_NONE;
+}
 //-------------------------------------------
 // make a buffer from an stext_page's text
 //-------------------------------------------
@@ -4409,3 +4466,5 @@ PyObject *pixmap_pixel(fz_pixmap* pm, int x, int y);
 int pixmap_n(mupdf::FzPixmap& pixmap);
 
 PyObject* JM_search_stext_page(fz_stext_page *page, const char *needle);
+
+PyObject *set_pixel(fz_pixmap* pm, int x, int y, PyObject *color);

tests/test_pixmap.py

@@ -152,3 +152,30 @@ def test_3020():
     pm2 = fitz.Pixmap(pm, 20, 30, None)
     pm3 = fitz.Pixmap(fitz.csGRAY, pm)
     pm4 = fitz.Pixmap(pm, pm3)
+
+def test_3050():
+    pdf_file = fitz.open(pdf)
+    for page_no, page in enumerate(pdf_file):
+        zoom_x = 4.0
+        zoom_y = 4.0
+        matrix = fitz.Matrix(zoom_x, zoom_y)
+        pix = page.get_pixmap(matrix=matrix)
+        digest0 = pix.digest
+        print(f'{pix.width=} {pix.height=}')
+        def product(x, y):
+            for yy in y:
+                for xx in x:
+                    yield (xx, yy)
+        n = 0
+        # We use a small subset of the image because non-optimised rebase gets
+        # very slow.
+        for pos in product(range(100), range(100)):
+            if sum(pix.pixel(pos[0], pos[1])) >= 600:
+                n += 1
+                pix.set_pixel(pos[0], pos[1], (255, 255, 255))
+        digest1 = pix.digest
+        print(f'{page_no=} {n=} {digest0=} {digest1=}')
+        digest_expected = b'\xd7x\x94_\x98\xa1<-/\xf3\xf9\x04\xec#\xaa\xee'
+        pix.save(os.path.abspath(f'{__file__}/../../tests/test_3050_out.png'))
+        assert digest1 != digest0
+        assert digest1 == digest_expected