TalkPipe 1.1.0__tar.gz → 1.2.0__tar.gz

{talkpipe-1.1.0 → talkpipe-1.2.0}/.pre-commit-config.yaml

@@ -2,7 +2,7 @@
 # per clone; `pre-commit run --all-files` reproduces the gate locally.
 repos:
   - repo: https://github.com/astral-sh/ruff-pre-commit
-    rev: v0.16.3  # keep in step with the ruff pin in pyproject.toml
+    rev: v0.17.0  # keep in step with the ruff pin in pyproject.toml
     hooks:
       - id: ruff-check
         args: [--fix, --show-fixes]

{talkpipe-1.1.0 → talkpipe-1.2.0}/CHANGELOG.md

@@ -2,6 +2,59 @@
 
 ## Unreleased
 
+## 1.2.0 (2026-10-24)
+
+- `build_rag_database` accepts pre-made chunks: a new keyword-only
+  `chunks=` argument (an iterable of dicts shaped like
+  `ProcessDocumentsSegment` output) is embedded and stored directly, with
+  `source_pattern` passed as `None`. The preflight, dimension check,
+  overflow truncation and skip accounting apply unchanged. This lets a
+  caller chunk documents once, index the chunks for keyword search, and
+  embed the same chunks later. Passing both or neither raises
+  `ValueError`; existing calls are unaffected.
+- **Security: raised the `Pillow` floor** in the `pillow` extra (and so in
+  `all`) to `Pillow>=13.0.0`, clearing six advisories against earlier
+  releases. `uv.lock` moves `pillow` to 13.0.0.
+- **Security: raised dependency floors** to clear advisories against
+  earlier releases: `pypdf>=6.21.2` in the `pypdf` extra (and so in `all`);
+  `certifi>=2026.10.19` in the core dependencies (it arrives through
+  `requests` and `httpx`); `filelock>=4.2.1` in the `model2vec` extra (it
+  arrives through `huggingface-hub`); and `virtualenv>=21.14.9` and
+  `filelock>=4.2.1` in the `dev` extra (they arrive through `pre-commit`).
+  `uv.lock` moves `certifi` to 2026.10.23, `filelock` to 4.2.4, `pypdf` to
+  6.21.2, `virtualenv` to 21.14.13, and `packaging` to 26.3.
+
+## 1.1.1 (2026-10-16)
+
+- A bare word in a chatterlang parameter that names no `CONST` (for
+  example `tools=[lookup_ticket]` instead of `tools=["lookup_ticket"]`)
+  now fails compilation with `Segment 'llmAgent': unknown constant
+  'lookup_ticket' in tools=. A bare word refers to a CONST; quote it for
+  the string "lookup_ticket", or declare it first (CONST lookup_ticket =
+  ...)`. It used to surface as the raw `KeyError` text, `Segment
+  'llmAgent' could not be created: 'lookup_ticket'`. The error is a
+  `bad_param` `CompileError` naming the segment or source, so the workbench
+  lint shows it like other parameter errors.
+- `--logger_levels` (and `configure_logger` / the `configureLogger`
+  segment) no longer prints every log line twice. A named logger got its
+  own console handler and its records also propagated to the handler
+  `logging.basicConfig` had put on the root logger; the named logger now
+  stops propagating. A `logger_files` entry for the root logger therefore
+  does not receive a named logger's records: name the same logger in
+  `logger_files` (the docstring says so).
+- `chatterlang_reference_generator` no longer drops the first parameter of
+  every plain segment from the `.html` and `.txt` references it writes.
+  The extractor already removes the data parameter (`items` / `item`) by
+  name, and the generator then removed the first remaining parameter by
+  position as well, so `cast` was documented without `cast_type`,
+  `llmPrompt` without `model`, `print` without `pprint`, and so on.
+  `chatterlang_reference_browser` and the workbench, which do not apply
+  that second filter, were already correct.
+- The `dev` extra moves ruff to the 0.17 series (`>=0.17.0,<0.18`), with
+  the ruff-pre-commit hook on v0.17.0 to match; a dependency update had
+  widened the pin to `<0.18` without moving the hook. The code needed no
+  changes for the new release's rules or formatting.
+
 ## 1.1.0 (2026-10-12)
 
 - Added `talkpipe.llm.config.resolve_source_and_model(source, model)`: applies

{talkpipe-1.1.0 → talkpipe-1.2.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: talkpipe
-Version: 1.1.0
+Version: 1.2.0
 Summary: Python internal and external DSL for writing generative AI analytics
 Author-email: Travis Bauer <tlbauer@sandia.gov>
 Maintainer-email: Travis Bauer <tlbauer@sandia.gov>
@@ -41,14 +41,17 @@ Requires-Dist: pyyaml>=5.4
 Requires-Dist: greenlet
 Requires-Dist: networkx
 Requires-Dist: starlette>=1.3.1
+Requires-Dist: certifi>=2026.10.19
 Provides-Extra: dev
 Requires-Dist: pytest>=9.0.3; extra == "dev"
 Requires-Dist: pytest-mock; extra == "dev"
 Requires-Dist: pytest-cov>=2.0.0; extra == "dev"
 Requires-Dist: mongomock; extra == "dev"
-Requires-Dist: ruff<0.18,>=0.16.3; extra == "dev"
+Requires-Dist: ruff<0.18,>=0.17.0; extra == "dev"
 Requires-Dist: mypy<3,>=2.3; extra == "dev"
 Requires-Dist: pre-commit>=4.0; extra == "dev"
+Requires-Dist: virtualenv>=21.14.9; extra == "dev"
+Requires-Dist: filelock>=4.2.1; extra == "dev"
 Provides-Extra: ollama
 Requires-Dist: ollama; extra == "ollama"
 Provides-Extra: openai
@@ -56,11 +59,12 @@ Requires-Dist: openai; extra == "openai"
 Provides-Extra: anthropic
 Requires-Dist: anthropic>=0.87.0; extra == "anthropic"
 Provides-Extra: pypdf
-Requires-Dist: pypdf>=6.19.0; extra == "pypdf"
+Requires-Dist: pypdf>=6.21.2; extra == "pypdf"
 Provides-Extra: pillow
-Requires-Dist: Pillow>=12.2.0; extra == "pillow"
+Requires-Dist: Pillow>=13.0.0; extra == "pillow"
 Provides-Extra: model2vec
 Requires-Dist: model2vec>=0.3.4; extra == "model2vec"
+Requires-Dist: filelock>=4.2.1; extra == "model2vec"
 Provides-Extra: all
 Requires-Dist: talkpipe[anthropic,model2vec,ollama,openai,pillow,pypdf]; extra == "all"
 Dynamic: license-file

{talkpipe-1.1.0 → talkpipe-1.2.0}/pyproject.toml

@@ -60,6 +60,10 @@ dependencies = [
     # covers the advisories published since, the latest being CVE-2026-54283,
     # which is first fixed in 1.3.1.
     "starlette>=1.3.1",
+    # Security override: certifi is a transitive dependency of requests and
+    # httpx, floored here because neither sets one that clears its
+    # advisories: <2026.10.19 has known advisories.
+    "certifi>=2026.10.19",
 ]
 dynamic = ["version"]
 
@@ -73,9 +77,14 @@ dev = [
     # new rules and mypy adds checks in minor releases, and CI resolves fresh,
     # so an open floor would turn a tool release into a red build. Bump the
     # pins deliberately (and the ruff-pre-commit rev with them).
-    "ruff>=0.16.3,<0.18",
+    "ruff>=0.17.0,<0.18",
     "mypy>=2.3,<3",
     "pre-commit>=4.0",
+    # Security overrides: virtualenv and filelock arrive through pre-commit,
+    # floored here because it sets no floor that clears their advisories
+    # (<21.14.9 and <4.2.1 respectively have known advisories).
+    "virtualenv>=21.14.9",
+    "filelock>=4.2.1",
 ]
 ollama = [
     'ollama'
@@ -87,13 +96,16 @@ anthropic = [
     'anthropic>=0.87.0'
 ]
 pypdf = [
-    'pypdf>=6.19.0'  # security floor: <6.19.0 has several crafted-PDF runtime and memory exhaustion advisories
+    'pypdf>=6.21.2'  # security floor: <6.21.2 has several crafted-PDF runtime and memory exhaustion advisories
 ]
 pillow = [
-    'Pillow>=12.2.0'
+    'Pillow>=13.0.0'  # security floor: <13.0.0 has six known advisories
 ]
 model2vec = [
     'model2vec>=0.3.4',
+    # Security override: filelock arrives through huggingface-hub, which sets
+    # no floor that clears its advisories (<4.2.1 has known advisories).
+    'filelock>=4.2.1',
 ]
 all = [
     'talkpipe[openai,ollama,anthropic,pypdf,pillow,model2vec]'

{talkpipe-1.1.0 → talkpipe-1.2.0}/src/talkpipe/app/chatterlang_reference_generator.py

@@ -177,24 +177,23 @@ def _display_type(item: AnalyzedItem) -> str:
 
 
 def _visible_params(item: AnalyzedItem, skip: list[str]) -> list[ParamSpec]:
-    """Parameters to document: drop the ``skip`` names, then drop the leading
-    data parameter for plain segments.
+    """Parameters to document: the item's parameters without the ``skip``
+    names.
 
     The two renderers skip different names, so the caller supplies the list.
+    The data parameter (``items`` / ``item``) is already gone by the time an
+    item gets here, because ``extract_component_info`` skips it by name, so
+    nothing is dropped by position: doing so removed the first real
+    parameter of every plain segment (``cast``'s ``cast_type``,
+    ``llmPrompt``'s ``model``).
     """
-    filtered_params = [p for p in item.parameters if p.name not in skip]
-    if item.is_segment and len(filtered_params) > 0:
-        # For segments, the first param is often the data being processed
-        filtered_params = (
-            filtered_params[1:] if len(filtered_params) > 1 else filtered_params
-        )
-    return filtered_params
+    return [p for p in item.parameters if p.name not in skip]
 
 
 def generate_html(analyzed_items: list[AnalyzedItem], output_file: str) -> None:
     """
     Generate a styled HTML file with a table of contents, raw HTML docstrings,
-    skipping 'self', and skipping first param for segment functions.
+    skipping 'self', 'items' and 'item'.
     """
     # Group by type (Source/Segment), then sort by chatterlang name
     items_by_type = _group_by_type(analyzed_items)
@@ -368,7 +367,7 @@ def generate_html(analyzed_items: list[AnalyzedItem], output_file: str) -> None:
                     f"  <div><pre>{html.escape(item.docstring)}</pre></div>\n"
                 )
 
-            # Process parameters - skip 'self' and for segments skip first param if it's the data param
+            # Process parameters - skip 'self' and the data parameter names
             filtered_params = _visible_params(item, ["self", "items", "item"])
 
             if filtered_params:
@@ -403,7 +402,7 @@ def generate_html(analyzed_items: list[AnalyzedItem], output_file: str) -> None:
 def generate_text(analyzed_items: list[AnalyzedItem], output_file: str) -> None:
     """
     Write a nicely formatted plain-text version of the same documentation.
-    Group by type, sorted, omit 'self', skip first param for segment functions.
+    Group by type, sorted, omit 'self'.
     No table of contents here - just a linear, grouped listing.
     """
     # Group by type (Source/Segment), then sort by chatterlang name

{talkpipe-1.1.0 → talkpipe-1.2.0}/src/talkpipe/chatterlang/compiler.py

@@ -473,20 +473,49 @@ def compile(
     return Script(final_pipelines)
 
 
-def _resolve_value(value: Any, runtime: RuntimeComponent) -> Any:
+def _unknown_constant_message(kind: str, name: str, param: str, word: str) -> str:
+    """Build a CompileError message for a bare word that names no constant.
+
+    A bare word in a parameter value is a reference to a ``CONST``; a
+    newcomer who writes ``tools=[lookup_ticket]`` meaning the string gets
+    ``KeyError: 'lookup_ticket'`` otherwise, which says nothing about quoting.
+    """
+    return (
+        f"{kind} '{name}': unknown constant '{word}' in {param}=. A bare word "
+        f'refers to a CONST; quote it for the string "{word}", or declare it '
+        f"first (CONST {word} = ...)"
+    )
+
+
+def _resolve_value(
+    value: Any, runtime: RuntimeComponent, *, kind: str, name: str, param: str
+) -> Any:
     """Resolve a single parameter value, handling constants and arrays recursively."""
     if isinstance(value, Identifier):
-        return runtime.const_store[value.name]
+        try:
+            return runtime.const_store[value.name]
+        except KeyError:
+            raise CompileError(
+                _unknown_constant_message(kind, name, param, value.name),
+                kind="bad_param",
+                bad_name=name,
+            ) from None
     if isinstance(value, list):
-        return [_resolve_value(elem, runtime) for elem in value]
+        return [
+            _resolve_value(elem, runtime, kind=kind, name=name, param=param)
+            for elem in value
+        ]
     return value
 
 
 def _resolve_params(
-    params: dict[Any, Any], runtime: RuntimeComponent
+    params: dict[Any, Any], runtime: RuntimeComponent, *, kind: str, name: str
 ) -> dict[str, Any]:
-    """Resolve the parameters for a segment"""
-    return {k: _resolve_value(params[k], runtime) for k in params}
+    """Resolve the parameters for the *kind* (Source / Segment) named *name*."""
+    return {
+        k: _resolve_value(params[k], runtime, kind=kind, name=name, param=k)
+        for k in params
+    }
 
 
 @compile.register(ParsedPipeline)
@@ -522,7 +551,12 @@ def _(
                 ) from None
             try:
                 ans = source_cls(
-                    **_resolve_params(pipeline.input_node.params, runtime=runtime)
+                    **_resolve_params(
+                        pipeline.input_node.params,
+                        runtime=runtime,
+                        kind="Source",
+                        name=source_name,
+                    )
                 )
             except TypeError as e:
                 raise CompileError(
@@ -560,7 +594,12 @@ def _(
                 ) from None
             try:
                 next_transform = segment_cls(
-                    **_resolve_params(transform.params, runtime=runtime)
+                    **_resolve_params(
+                        transform.params,
+                        runtime=runtime,
+                        kind="Segment",
+                        name=segment_name,
+                    )
                 )
             except TypeError as e:
                 raise CompileError(

{talkpipe-1.1.0 → talkpipe-1.2.0}/src/talkpipe/pipelines/vector_databases.py

@@ -348,7 +348,7 @@ def _tally_stored_chunks(
 
 
 def build_rag_database(
-    source_pattern: str | Iterable[str],
+    source_pattern: str | Iterable[str] | None,
     path: str,
     embedding_model: str | None = None,
     embedding_source: str | None = None,
@@ -366,6 +366,7 @@ def build_rag_database(
     expected_dimension: int | None = None,
     preflight: bool = True,
     progress: Callable[[int, int, str], None] | None = None,
+    chunks: Iterable[dict[str, Any]] | None = None,
 ) -> RagIngestResult:
     """Build a RAG vector database from documents matching a glob pattern.
 
@@ -393,7 +394,16 @@ def build_rag_database(
 
     The optional progress callback receives (chunks_done, files_done,
     current_source_path) as chunks are stored.
+
+    chunks, when given, replaces reading and chunking: the dicts (already
+    shaped like ProcessDocumentsSegment output, with embedding_field and a
+    "source" field) are embedded and stored directly, and source_pattern
+    must be None. Callers that chunk documents themselves, for example to
+    build a keyword index from the same chunks first, keep the preflight,
+    dimension check, overflow handling, and skip accounting above.
     """
+    if (chunks is None) == (source_pattern is None):
+        raise ValueError("pass exactly one of source_pattern and chunks")
     config = get_config()
     embedding_model = embedding_model or config.get(TALKPIPE_EMBEDDING_MODEL_NAME)
     embedding_source = embedding_source or config.get(TALKPIPE_EMBEDDING_MODEL_SOURCE)
@@ -431,14 +441,18 @@ def build_rag_database(
             )
 
     tally = _IngestTally()
-    pipeline = (
-        ProcessDocumentsSegment(
-            chunk_size=chunk_size,
-            shingle_size=shingle_size,
-            overlap=overlap,
+    tally_and_embed = _tally_extracted_chunks(tally=tally) | embed_segment
+    if chunks is None:
+        tally_and_embed = (
+            ProcessDocumentsSegment(
+                chunk_size=chunk_size,
+                shingle_size=shingle_size,
+                overlap=overlap,
+            )
+            | tally_and_embed
         )
-        | _tally_extracted_chunks(tally=tally)
-        | embed_segment
+    pipeline = (
+        tally_and_embed
         | add_to_lancedb(
             path=path,
             table_name=table_name,
@@ -449,10 +463,13 @@ def build_rag_database(
         | _tally_stored_chunks(tally=tally, progress=progress)
     )
 
-    patterns = (
-        [source_pattern] if isinstance(source_pattern, str) else list(source_pattern)
-    )
-    for _ in pipeline.transform(patterns):
+    if chunks is not None:
+        inputs: Iterable[Any] = chunks
+    elif isinstance(source_pattern, str):
+        inputs = [source_pattern]
+    else:
+        inputs = list(source_pattern or [])
+    for _ in pipeline.transform(inputs):
         pass
     # Chunks consumed after the last stored one also failed to embed.
     tally.chunks_skipped += tally.chunks_extracted - tally.extracted_at_last_store

{talkpipe-1.1.0 → talkpipe-1.2.0}/src/talkpipe/util/config.py

@@ -257,6 +257,10 @@ def configure_logger(
         This reconfigures the calling process's logging: it calls
         ``logging.basicConfig`` and clears the existing handlers of every
         logger named in ``logger_levels`` before attaching a console handler.
+        A named (non-root) logger then stops propagating to the root
+        logger, so each record is printed once; a ``logger_files`` entry
+        for the root logger therefore does not receive that logger's
+        records (name the same logger in ``logger_files``).
         Applications embedding talkpipe that manage their own logging should
         not call it (or use the ``configureLogger`` segment).
     """
@@ -289,6 +293,10 @@ def configure_logger(
             console_handler.setLevel(level)
             console_handler.setFormatter(formatter)
             logger.addHandler(console_handler)
+            # The console handler above is this logger's output; without
+            # this, every record also propagated to the root handler that
+            # basicConfig installed and printed a second time.
+            logger.propagate = logger_name == "root"
 
     if logger_files:
         for logger_name, file_name in parse_key_value_str(logger_files).items():

{talkpipe-1.1.0 → talkpipe-1.2.0}/src/talkpipe.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: talkpipe
-Version: 1.1.0
+Version: 1.2.0
 Summary: Python internal and external DSL for writing generative AI analytics
 Author-email: Travis Bauer <tlbauer@sandia.gov>
 Maintainer-email: Travis Bauer <tlbauer@sandia.gov>
@@ -41,14 +41,17 @@ Requires-Dist: pyyaml>=5.4
 Requires-Dist: greenlet
 Requires-Dist: networkx
 Requires-Dist: starlette>=1.3.1
+Requires-Dist: certifi>=2026.10.19
 Provides-Extra: dev
 Requires-Dist: pytest>=9.0.3; extra == "dev"
 Requires-Dist: pytest-mock; extra == "dev"
 Requires-Dist: pytest-cov>=2.0.0; extra == "dev"
 Requires-Dist: mongomock; extra == "dev"
-Requires-Dist: ruff<0.18,>=0.16.3; extra == "dev"
+Requires-Dist: ruff<0.18,>=0.17.0; extra == "dev"
 Requires-Dist: mypy<3,>=2.3; extra == "dev"
 Requires-Dist: pre-commit>=4.0; extra == "dev"
+Requires-Dist: virtualenv>=21.14.9; extra == "dev"
+Requires-Dist: filelock>=4.2.1; extra == "dev"
 Provides-Extra: ollama
 Requires-Dist: ollama; extra == "ollama"
 Provides-Extra: openai
@@ -56,11 +59,12 @@ Requires-Dist: openai; extra == "openai"
 Provides-Extra: anthropic
 Requires-Dist: anthropic>=0.87.0; extra == "anthropic"
 Provides-Extra: pypdf
-Requires-Dist: pypdf>=6.19.0; extra == "pypdf"
+Requires-Dist: pypdf>=6.21.2; extra == "pypdf"
 Provides-Extra: pillow
-Requires-Dist: Pillow>=12.2.0; extra == "pillow"
+Requires-Dist: Pillow>=13.0.0; extra == "pillow"
 Provides-Extra: model2vec
 Requires-Dist: model2vec>=0.3.4; extra == "model2vec"
+Requires-Dist: filelock>=4.2.1; extra == "model2vec"
 Provides-Extra: all
 Requires-Dist: talkpipe[anthropic,model2vec,ollama,openai,pillow,pypdf]; extra == "all"
 Dynamic: license-file

{talkpipe-1.1.0 → talkpipe-1.2.0}/src/talkpipe.egg-info/requires.txt

@@ -19,6 +19,7 @@ pyyaml>=5.4
 greenlet
 networkx
 starlette>=1.3.1
+certifi>=2026.10.19
 
 [all]
 talkpipe[anthropic,model2vec,ollama,openai,pillow,pypdf]
@@ -31,12 +32,15 @@ pytest>=9.0.3
 pytest-mock
 pytest-cov>=2.0.0
 mongomock
-ruff<0.18,>=0.16.3
+ruff<0.18,>=0.17.0
 mypy<3,>=2.3
 pre-commit>=4.0
+virtualenv>=21.14.9
+filelock>=4.2.1
 
 [model2vec]
 model2vec>=0.3.4
+filelock>=4.2.1
 
 [ollama]
 ollama
@@ -45,7 +49,7 @@ ollama
 openai
 
 [pillow]
-Pillow>=12.2.0
+Pillow>=13.0.0
 
 [pypdf]
-pypdf>=6.19.0
+pypdf>=6.21.2

talkpipe-1.2.0/src/talkpipe.egg-info/scm_version.json

@@ -0,0 +1,8 @@
+{
+  "tag": "1.2.0",
+  "distance": 0,
+  "node": "g5833aeff5bb8a583ea48888a7ee283f541c3c28c",
+  "dirty": false,
+  "branch": "HEAD",
+  "node_date": "2026-10-24"
+}

{talkpipe-1.1.0 → talkpipe-1.2.0}/tests/talkpipe/app/test_chatterlang_reference_generator.py

@@ -250,6 +250,36 @@ class TestGeneration:
         finally:
             os.unlink(temp_file)
 
+    def test_segments_keep_their_first_parameter(self):
+        """Every parameter the extractor returns is rendered. The data
+        parameter is already skipped by name upstream, so dropping the first
+        remaining one by position removed a real parameter from every plain
+        segment (``cast`` was documented without ``cast_type``)."""
+        item = AnalyzedItem(
+            type="Segment",
+            name="Cast",
+            docstring="Casts the input data.",
+            parameters=[
+                ParamSpec("cast_type", "str"),
+                ParamSpec("fail_silently", "bool", "True"),
+            ],
+            chatterlang_name="cast",
+            decorator_type="segment",
+            is_segment=True,
+        )
+        with tempfile.TemporaryDirectory() as tmp:
+            html_file = os.path.join(tmp, "ref.html")
+            text_file = os.path.join(tmp, "ref.txt")
+            generate_html([item], html_file)
+            generate_text([item], text_file)
+            with open(html_file, encoding="utf-8") as f:
+                html_content = f.read()
+            with open(text_file, encoding="utf-8") as f:
+                text_content = f.read()
+        for content in (html_content, text_content):
+            assert "cast_type" in content
+            assert "fail_silently" in content
+
 
 class TestMainFunctions:
     """Test main entry points."""

{talkpipe-1.1.0 → talkpipe-1.2.0}/tests/talkpipe/chatterlang/test_compiler.py

@@ -593,6 +593,31 @@ def test_compile_error_when_component_rejects_its_options():
     assert excinfo.value.__suppress_context__
 
 
+def test_compile_error_names_an_unknown_constant():
+    """A bare word in a parameter is a CONST reference. Unknown, it used to
+    surface as the KeyError text (``could not be created: 'foo'``), which
+    says nothing about quoting or CONST."""
+    with pytest.raises(compiler.CompileError) as excinfo:
+        compiler.compile('INPUT FROM echo[data="a,b"] | firstN[n=[foo]]')
+    msg = str(excinfo.value)
+    assert "Segment 'firstN': unknown constant 'foo' in n=" in msg
+    assert 'quote it for the string "foo"' in msg
+    assert "CONST foo = ..." in msg
+    assert excinfo.value.kind == "bad_param"
+    assert excinfo.value.bad_name == "firstN"
+    assert excinfo.value.__cause__ is None
+
+    with pytest.raises(compiler.CompileError) as excinfo:
+        compiler.compile("INPUT FROM echo[data=foo] | print")
+    assert "Source 'echo': unknown constant 'foo' in data=" in str(excinfo.value)
+
+    # A declared constant still resolves, in a list as well as bare.
+    script = compiler.compile(
+        'CONST foo = 1; INPUT FROM echo[data="a,b"] | firstN[n=foo] | firstN[n=[foo]]'
+    )
+    assert script is not None
+
+
 def test_compile_error_when_source_rejects_its_options():
     """The same wrapping applies to a source constructor."""
 

{talkpipe-1.1.0 → talkpipe-1.2.0}/tests/talkpipe/pipelines/test_vector_databases.py

@@ -966,6 +966,48 @@ def test_build_rag_database_raises_when_nothing_embeds(
         )
 
 
+def test_build_rag_database_accepts_premade_chunks(tmp_path, monkeypatch):
+    from talkpipe.pipelines import vector_databases as vdb
+
+    monkeypatch.setattr(vdb, "LLMEmbed", _fake_llm_embed(drop_indices=(1,)))
+    db_path = str(tmp_path / "db")
+    chunks = [
+        {
+            "id": f"c{n}",
+            "content": f"text {n}",
+            "shingle_text": f"text {n}",
+            "source": f"/docs/{n % 2}.txt",
+            "title": f"{n % 2}.txt",
+        }
+        for n in range(3)
+    ]
+
+    result = vdb.build_rag_database(
+        None,
+        path=db_path,
+        embedding_model="fake-model",
+        embedding_source="fake-source",
+        doc_id_field="id",
+        chunks=chunks,
+    )
+
+    assert result.chunks_indexed == 2
+    assert result.chunks_skipped == 1
+    assert result.files_indexed == 1
+    store = LanceDBDocumentStore(path=db_path, table_name="docs", vector_dim=3)
+    assert store.count() == 2
+    assert store.get_document("c2") is not None
+
+
+def test_build_rag_database_requires_exactly_one_input(tmp_path, rag_corpus):
+    from talkpipe.pipelines import vector_databases as vdb
+
+    with pytest.raises(ValueError, match="exactly one"):
+        vdb.build_rag_database(None, path=str(tmp_path / "db"))
+    with pytest.raises(ValueError, match="exactly one"):
+        vdb.build_rag_database(rag_corpus, path=str(tmp_path / "db"), chunks=[])
+
+
 def test_make_vector_database_passes_on_token_overflow(monkeypatch, tmp_path):
     from talkpipe.pipelines import vector_databases as vdb
 

{talkpipe-1.1.0 → talkpipe-1.2.0}/tests/talkpipe/util/test_util.py

@@ -529,6 +529,20 @@ def test_configure_logging_files_only(tmp_path):
     assert "files-only debug message" not in contents
 
 
+def test_configure_logging_prints_each_record_once(capsys):
+    """``configure_logger("name:LEVEL")`` gives the named logger its own
+    console handler; the record used to propagate to the root handler that
+    ``basicConfig`` installed as well, so every line appeared twice."""
+    try:
+        talkpipe.util.config.configure_logger("talkpipe.test.once:INFO")
+        logger = logging.getLogger("talkpipe.test.once")
+        logger.info("printed exactly once")
+        captured = capsys.readouterr()
+        assert captured.err.count("printed exactly once") == 1
+    finally:
+        logging.getLogger("talkpipe.test.once").propagate = True
+
+
 def test_talkpipe_logger_has_null_handler():
     import logging