Fix llama.cpp prebuilt: skip already-installed same-release fallback (#6285)

* Fix llama.cpp prebuilt: skip the already-installed same-release fallback

install_prebuilt computes diffusion_visual_server_backfill_needed from the
newest candidate (plan.attempts[0]); when that is True it passed
existing_install_dir=None to validate_prebuilt_attempts, which disabled the
"existing install already matches this candidate" skip for the WHOLE plan. So
when the newest bundle failed validation the installer re-downloaded and
re-extracted an older fallback bundle that was already correctly installed.

Pass the real install dir always and gate the skip per-attempt: a matching
candidate is skipped unless that specific candidate still needs the
DiffusionGemma backfill re-extract.

Also make test_llama_cpp_search_roots_handles_studio_root_oserror read the full
_find_llama_server_binary / _kill_orphaned_servers method bodies instead of a
fixed 4000-char window. The except handler it asserts already exists, but the
function grew past the window so the guard silently failed; slicing to the next
sibling def keeps the check correct as the file grows.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

---------

Co-authored-by: Daniel Han <michaelhan2050@gmail.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

studio/install_llama_prebuilt.py

@@ -6436,13 +6436,19 @@ def validate_prebuilt_attempts(
                 f"runtime_line={attempt.runtime_line} coverage_class={attempt.coverage_class}"
             )
 
-        if existing_install_dir is not None and existing_install_matches_choice(
-            existing_install_dir,
-            host,
-            llama_tag = llama_tag,
-            release_tag = release_tag,
-            choice = attempt,
-            approved_checksums = approved_checksums,
+        if (
+            existing_install_dir is not None
+            and existing_install_matches_choice(
+                existing_install_dir,
+                host,
+                llama_tag = llama_tag,
+                release_tag = release_tag,
+                choice = attempt,
+                approved_checksums = approved_checksums,
+            )
+            # Skip a matching candidate unless it still needs the DiffusionGemma
+            # backfill re-extract (gated per-attempt, not per-plan).
+            and not diffusion_visual_server_backfill_needed(existing_install_dir, host, attempt)
         ):
             log(
                 "existing llama.cpp install already matches fallback candidate "
@@ -6601,9 +6607,8 @@ def install_prebuilt(
                             release_tag = plan.release_tag,
                             approved_checksums = plan.approved_checksums,
                             initial_fallback_used = release_index > 0,
-                            # a backfill must reinstall, so do not let the inner
-                            # existing-install match short-circuit the re-extract
-                            existing_install_dir = None if backfill else install_dir,
+                            # Skip is gated per-attempt inside, so pass the dir always.
+                            existing_install_dir = install_dir,
                         )
                     except ExistingInstallSatisfied:
                         return

tests/test_studio_install_workspace_guard.py

@@ -873,15 +873,20 @@ def test_llama_cpp_search_roots_handles_studio_root_oserror():
     llama_cpp = (
         REPO_ROOT / "studio" / "backend" / "core" / "inference" / "llama_cpp.py"
     ).read_text()
-    find_block_start = llama_cpp.index("_find_llama_server_binary")
-    find_block = llama_cpp[find_block_start : find_block_start + 4000]
-    assert (
-        "except (ImportError, OSError, ValueError):" in find_block
+
+    def _method_body(name: str) -> str:
+        # Whole method body (def to next sibling def), so the check survives the
+        # function growing past any fixed-size window.
+        start = llama_cpp.index(f"def {name}")
+        indent = " " * (start - llama_cpp.rfind("\n", 0, start) - 1)
+        nxt = llama_cpp.find(f"\n{indent}def ", start + 1)
+        return llama_cpp[start : nxt if nxt != -1 else len(llama_cpp)]
+
+    assert "except (ImportError, OSError, ValueError):" in _method_body(
+        "_find_llama_server_binary"
     ), "_find_llama_server_binary must catch (ImportError, OSError, ValueError) from studio_root()"
-    kill_def_idx = llama_cpp.index("def _kill_orphaned_servers")
-    kill_block = llama_cpp[kill_def_idx : kill_def_idx + 4000]
-    assert (
-        "except (ImportError, OSError, ValueError):" in kill_block
+    assert "except (ImportError, OSError, ValueError):" in _method_body(
+        "_kill_orphaned_servers"
     ), "sibling _kill_orphaned_servers must keep its (ImportError, OSError, ValueError) handler"