git-commit-message 0.8.2__tar.gz → 0.9.0__tar.gz

{git_commit_message-0.8.2 → git_commit_message-0.9.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: git-commit-message
-Version: 0.8.2
+Version: 0.9.0
 Summary: Generate Git commit messages from staged changes using LLM
 Maintainer-email: Mina Her <minacle@live.com>
 License: This is free and unencumbered software released into the public domain.
@@ -47,7 +47,6 @@ Requires-Dist: babel>=2.17.0
 Requires-Dist: google-genai>=1.56.0
 Requires-Dist: ollama>=0.4.0
 Requires-Dist: openai>=2.6.1
-Requires-Dist: tiktoken>=0.12.0
 
 # git-commit-message
 
@@ -167,6 +166,18 @@ git-commit-message --one-line "optional context"
 git-commit-message --one-line --co-author 'John Doe <john.doe@example.com>'
 ```
 
+Use Conventional Commits constraints for the subject/footer only (body format is preserved):
+
+```sh
+git-commit-message --conventional
+
+# can be combined with one-line mode
+git-commit-message --conventional --one-line
+
+# co-author trailers are appended after any existing footers
+git-commit-message --conventional --co-author copilot
+```
+
 Select provider:
 
 ```sh
@@ -223,10 +234,24 @@ git-commit-message --chunk-tokens 0
 # chunk the diff into ~4000-token pieces before summarising
 git-commit-message --chunk-tokens 4000
 
+# note: for provider 'ollama', values >= 1 are not supported
+# use 0 (single summary pass) or -1 (legacy one-shot)
+git-commit-message --provider ollama --chunk-tokens 0
+
 # disable summarisation and use the legacy one-shot prompt
 git-commit-message --chunk-tokens -1
 ```
 
+Adjust unified diff context lines:
+
+```sh
+# use 5 context lines around each change hunk
+git-commit-message --diff-context 5
+
+# include only changed lines (no surrounding context)
+git-commit-message --diff-context 0
+```
+
 Select output language/locale (IETF language tag):
 
 ```sh
@@ -258,9 +283,11 @@ git-commit-message --provider llamacpp --host http://192.168.1.100:8080
 - `--provider {openai,google,ollama,llamacpp}`: provider to use (default: `openai`)
 - `--model MODEL`: model override (provider-specific; ignored for llama.cpp)
 - `--language TAG`: output language/locale (default: `en-GB`)
+- `--conventional`: apply Conventional Commits constraints to the subject and footer behavior. The body format is unchanged and still includes the translated `Rationale:` line. Breaking changes are expressed with `!` in the subject line, and `BREAKING CHANGE` footer lines are not generated.
 - `--one-line`: output subject only when no trailers are appended; with `--co-author`, output is a single-line subject plus `Co-authored-by:` trailer lines
 - `--max-length N`: max subject length (default: 72)
-- `--chunk-tokens N`: token budget per diff chunk (`0` = single summary pass, `-1` disables summarisation)
+- `--chunk-tokens N`: token budget per diff chunk (`0` = single summary pass, `-1` disables summarisation). For `ollama`, values `>= 1` are not supported.
+- `--diff-context N`: context lines in unified diff (`N >= 0`). If omitted, uses `GIT_COMMIT_MESSAGE_DIFF_CONTEXT` when set; otherwise uses Git default (usually `3`).
 - `--debug`: print request/response details
 - `--commit`: run `git commit -m <message>`
 - `--amend`: generate a message suitable for amending the previous commit (diff is from the amended commit's parent to the staged index; if nothing is staged, this effectively becomes the diff introduced by `HEAD`)
@@ -284,7 +311,8 @@ Optional:
 - `OLLAMA_HOST`: Ollama server URL (default: `http://localhost:11434`)
 - `LLAMACPP_HOST`: llama.cpp server URL (default: `http://localhost:8080`)
 - `GIT_COMMIT_MESSAGE_LANGUAGE`: default language/locale (default: `en-GB`)
-- `GIT_COMMIT_MESSAGE_CHUNK_TOKENS`: default chunk token budget (default: `0`)
+- `GIT_COMMIT_MESSAGE_CHUNK_TOKENS`: default chunk token budget (default: `0`; for `ollama`, values `>= 1` are not supported)
+- `GIT_COMMIT_MESSAGE_DIFF_CONTEXT`: default unified diff context lines (`0` or greater). If unset, Git default is used (usually `3`).
 
 Default models (if not overridden):
 

{git_commit_message-0.8.2 → git_commit_message-0.9.0}/README.md

@@ -116,6 +116,18 @@ git-commit-message --one-line "optional context"
 git-commit-message --one-line --co-author 'John Doe <john.doe@example.com>'
 ```
 
+Use Conventional Commits constraints for the subject/footer only (body format is preserved):
+
+```sh
+git-commit-message --conventional
+
+# can be combined with one-line mode
+git-commit-message --conventional --one-line
+
+# co-author trailers are appended after any existing footers
+git-commit-message --conventional --co-author copilot
+```
+
 Select provider:
 
 ```sh
@@ -172,10 +184,24 @@ git-commit-message --chunk-tokens 0
 # chunk the diff into ~4000-token pieces before summarising
 git-commit-message --chunk-tokens 4000
 
+# note: for provider 'ollama', values >= 1 are not supported
+# use 0 (single summary pass) or -1 (legacy one-shot)
+git-commit-message --provider ollama --chunk-tokens 0
+
 # disable summarisation and use the legacy one-shot prompt
 git-commit-message --chunk-tokens -1
 ```
 
+Adjust unified diff context lines:
+
+```sh
+# use 5 context lines around each change hunk
+git-commit-message --diff-context 5
+
+# include only changed lines (no surrounding context)
+git-commit-message --diff-context 0
+```
+
 Select output language/locale (IETF language tag):
 
 ```sh
@@ -207,9 +233,11 @@ git-commit-message --provider llamacpp --host http://192.168.1.100:8080
 - `--provider {openai,google,ollama,llamacpp}`: provider to use (default: `openai`)
 - `--model MODEL`: model override (provider-specific; ignored for llama.cpp)
 - `--language TAG`: output language/locale (default: `en-GB`)
+- `--conventional`: apply Conventional Commits constraints to the subject and footer behavior. The body format is unchanged and still includes the translated `Rationale:` line. Breaking changes are expressed with `!` in the subject line, and `BREAKING CHANGE` footer lines are not generated.
 - `--one-line`: output subject only when no trailers are appended; with `--co-author`, output is a single-line subject plus `Co-authored-by:` trailer lines
 - `--max-length N`: max subject length (default: 72)
-- `--chunk-tokens N`: token budget per diff chunk (`0` = single summary pass, `-1` disables summarisation)
+- `--chunk-tokens N`: token budget per diff chunk (`0` = single summary pass, `-1` disables summarisation). For `ollama`, values `>= 1` are not supported.
+- `--diff-context N`: context lines in unified diff (`N >= 0`). If omitted, uses `GIT_COMMIT_MESSAGE_DIFF_CONTEXT` when set; otherwise uses Git default (usually `3`).
 - `--debug`: print request/response details
 - `--commit`: run `git commit -m <message>`
 - `--amend`: generate a message suitable for amending the previous commit (diff is from the amended commit's parent to the staged index; if nothing is staged, this effectively becomes the diff introduced by `HEAD`)
@@ -233,7 +261,8 @@ Optional:
 - `OLLAMA_HOST`: Ollama server URL (default: `http://localhost:11434`)
 - `LLAMACPP_HOST`: llama.cpp server URL (default: `http://localhost:8080`)
 - `GIT_COMMIT_MESSAGE_LANGUAGE`: default language/locale (default: `en-GB`)
-- `GIT_COMMIT_MESSAGE_CHUNK_TOKENS`: default chunk token budget (default: `0`)
+- `GIT_COMMIT_MESSAGE_CHUNK_TOKENS`: default chunk token budget (default: `0`; for `ollama`, values `>= 1` are not supported)
+- `GIT_COMMIT_MESSAGE_DIFF_CONTEXT`: default unified diff context lines (`0` or greater). If unset, Git default is used (usually `3`).
 
 Default models (if not overridden):
 

{git_commit_message-0.8.2 → git_commit_message-0.9.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "git-commit-message"
-version = "0.8.2"
+version = "0.9.0"
 description = "Generate Git commit messages from staged changes using LLM"
 readme = "README.md"
 requires-python = ">=3.13"
@@ -9,7 +9,6 @@ dependencies = [
 	"google-genai>=1.56.0",
 	"ollama>=0.4.0",
 	"openai>=2.6.1",
-	"tiktoken>=0.12.0",
 ]
 maintainers = [{ name = "Mina Her", email = "minacle@live.com" }]
 license = { file = "UNLICENSE" }

{git_commit_message-0.8.2 → git_commit_message-0.9.0}/src/git_commit_message/_cli.py

@@ -23,6 +23,7 @@ from ._git import (
     has_staged_changes,
     resolve_amend_base_ref,
 )
+from ._config import resolve_provider_name, validate_provider_chunk_tokens
 from ._llm import (
     CommitMessageResult,
     UnsupportedProviderError,
@@ -37,6 +38,7 @@ class CliArgs(Namespace):
         "commit",
         "amend",
         "edit",
+        "conventional",
         "provider",
         "model",
         "language",
@@ -44,6 +46,7 @@ class CliArgs(Namespace):
         "one_line",
         "max_length",
         "chunk_tokens",
+        "diff_context",
         "host",
         "co_authors",
     )
@@ -56,6 +59,7 @@ class CliArgs(Namespace):
         self.commit: bool = False
         self.amend: bool = False
         self.edit: bool = False
+        self.conventional: bool = False
         self.provider: str | None = None
         self.model: str | None = None
         self.language: str | None = None
@@ -63,6 +67,7 @@ class CliArgs(Namespace):
         self.one_line: bool = False
         self.max_length: int | None = None
         self.chunk_tokens: int | None = None
+        self.diff_context: int | None = None
         self.host: str | None = None
         self.co_authors: list[str] | None = None
 
@@ -155,6 +160,21 @@ def _env_chunk_tokens_default() -> int | None:
         return None
 
 
+def _env_diff_context_default() -> int | None:
+    """Return diff context default from env.
+
+    Raises
+    ------
+    ValueError
+        If the configured value is not an integer.
+    """
+
+    raw: str | None = environ.get("GIT_COMMIT_MESSAGE_DIFF_CONTEXT")
+    if raw is None:
+        return None
+    return int(raw)
+
+
 def _build_parser() -> ArgumentParser:
     """Create the CLI argument parser.
 
@@ -199,6 +219,15 @@ def _build_parser() -> ArgumentParser:
         help="Open an editor to amend the message before committing. Use with '--commit'.",
     )
 
+    parser.add_argument(
+        "--conventional",
+        action="store_true",
+        help=(
+            "Use Conventional Commits constraints for the subject line and footer. "
+            "The existing body format remains unchanged, including the translated Rationale line."
+        ),
+    )
+
     parser.add_argument(
         "--provider",
         default=None,
@@ -258,10 +287,23 @@ def _build_parser() -> ArgumentParser:
         help=(
             "Target token budget per diff chunk. "
             "0 forces a single chunk with summarisation; -1 disables summarisation (legacy one-shot). "
+            "For provider 'ollama', values >= 1 are not supported. "
             "If omitted, uses GIT_COMMIT_MESSAGE_CHUNK_TOKENS when set (default: 0)."
         ),
     )
 
+    parser.add_argument(
+        "--diff-context",
+        dest="diff_context",
+        type=int,
+        default=None,
+        help=(
+            "Number of context lines in unified diff output. "
+            "If omitted, uses GIT_COMMIT_MESSAGE_DIFF_CONTEXT when set "
+            "(default: Git default, usually 3)."
+        ),
+    )
+
     parser.add_argument(
         "--host",
         dest="host",
@@ -308,6 +350,32 @@ def _run(
         Process exit code. 0 indicates success; any other value indicates failure.
     """
 
+    chunk_tokens: int | None = args.chunk_tokens
+    if chunk_tokens is None:
+        chunk_tokens = _env_chunk_tokens_default()
+    if chunk_tokens is None:
+        chunk_tokens = 0
+
+    diff_context: int | None = args.diff_context
+    if diff_context is None:
+        try:
+            diff_context = _env_diff_context_default()
+        except ValueError:
+            print(
+                "GIT_COMMIT_MESSAGE_DIFF_CONTEXT must be an integer.",
+                file=stderr,
+            )
+            return 2
+    if diff_context is not None and diff_context < 0:
+        print("--diff-context must be greater than or equal to 0.", file=stderr)
+        return 2
+
+    provider_name: str = resolve_provider_name(args.provider)
+    provider_arg_error = validate_provider_chunk_tokens(provider_name, chunk_tokens)
+    if provider_arg_error is not None:
+        print(provider_arg_error, file=stderr)
+        return 2
+
     repo_root: Path = get_repo_root()
 
     if args.amend:
@@ -316,22 +384,20 @@ def _run(
             return 2
 
         base_ref = resolve_amend_base_ref(repo_root)
-        diff_text: str = get_staged_diff(repo_root, base_ref=base_ref)
+        diff_text: str = get_staged_diff(
+            repo_root,
+            base_ref=base_ref,
+            context_lines=diff_context,
+        )
     else:
         if not has_staged_changes(repo_root):
             print("No staged changes. Run 'git add' and try again.", file=stderr)
             return 2
 
-        diff_text = get_staged_diff(repo_root)
+        diff_text = get_staged_diff(repo_root, context_lines=diff_context)
 
     hint: str | None = args.description if isinstance(args.description, str) else None
 
-    chunk_tokens: int | None = args.chunk_tokens
-    if chunk_tokens is None:
-        chunk_tokens = _env_chunk_tokens_default()
-    if chunk_tokens is None:
-        chunk_tokens = 0
-
     normalized_co_authors: list[str] | None = None
     if args.co_authors:
         try:
@@ -353,6 +419,7 @@ def _run(
                 chunk_tokens,
                 args.provider,
                 args.host,
+                args.conventional,
             )
             message = result.message
         else:
@@ -366,6 +433,7 @@ def _run(
                 chunk_tokens,
                 args.provider,
                 args.host,
+                args.conventional,
             )
     except UnsupportedProviderError as exc:
         print(str(exc), file=stderr)

git_commit_message-0.9.0/src/git_commit_message/_config.py

@@ -0,0 +1,71 @@
+"""Shared configuration resolvers for provider/model/language selection."""
+
+from __future__ import annotations
+
+from os import environ
+from typing import Final
+
+
+DEFAULT_PROVIDER: Final[str] = "openai"
+DEFAULT_MODEL_OPENAI: Final[str] = "gpt-5-mini"
+DEFAULT_MODEL_GOOGLE: Final[str] = "gemini-2.5-flash"
+DEFAULT_MODEL_OLLAMA: Final[str] = "gpt-oss:20b"
+DEFAULT_MODEL_LLAMACPP: Final[str] = "default"
+DEFAULT_LANGUAGE: Final[str] = "en-GB"
+
+
+def resolve_provider_name(
+    provider: str | None,
+    /,
+) -> str:
+    chosen = provider or environ.get("GIT_COMMIT_MESSAGE_PROVIDER") or DEFAULT_PROVIDER
+    return chosen.strip().lower()
+
+
+def resolve_model_name(
+    model: str | None,
+    provider_name: str,
+    /,
+) -> str:
+    if provider_name == "google":
+        default_model = DEFAULT_MODEL_GOOGLE
+        provider_model = None
+    elif provider_name == "ollama":
+        default_model = DEFAULT_MODEL_OLLAMA
+        provider_model = environ.get("OLLAMA_MODEL")
+    elif provider_name == "llamacpp":
+        default_model = DEFAULT_MODEL_LLAMACPP
+        provider_model = environ.get("LLAMACPP_MODEL")
+    else:
+        default_model = DEFAULT_MODEL_OPENAI
+        provider_model = environ.get("OPENAI_MODEL")
+
+    return model or environ.get("GIT_COMMIT_MESSAGE_MODEL") or provider_model or default_model
+
+
+def resolve_language_tag(
+    language: str | None,
+    /,
+) -> str:
+    return language or environ.get("GIT_COMMIT_MESSAGE_LANGUAGE") or DEFAULT_LANGUAGE
+
+
+def validate_provider_chunk_tokens(
+    provider_name: str,
+    chunk_tokens: int,
+    /,
+) -> str | None:
+    if chunk_tokens < -1:
+        return (
+            "'--chunk-tokens' must be -1 or greater. "
+            "Use -1 to disable summarisation, or 0/positive values to enable summarisation."
+        )
+
+    if provider_name == "ollama" and chunk_tokens > 0:
+        return (
+            "'--chunk-tokens' with values >= 1 is not supported for provider 'ollama'. "
+            "Use '--chunk-tokens 0' (single summary pass) or '--chunk-tokens -1' "
+            "(disable summarisation)."
+        )
+
+    return None

{git_commit_message-0.8.2 → git_commit_message-0.9.0}/src/git_commit_message/_git.py

@@ -183,6 +183,7 @@ def get_staged_diff(
     /,
     *,
     base_ref: str | None = None,
+    context_lines: int | None = None,
 ) -> str:
     """Return the staged changes as diff text.
 
@@ -195,6 +196,9 @@ def get_staged_diff(
         commit hash, or the empty tree hash) to diff against. When provided,
         the diff shows changes from ``base_ref`` to the staged index, instead
         of changes from ``HEAD`` to the staged index.
+    context_lines
+        Optional number of context lines for unified diff output. When ``None``,
+        Git's default context lines are used.
 
     Returns
     -------
@@ -210,6 +214,8 @@ def get_staged_diff(
         "--minimal",
         "--no-color",
     ]
+    if context_lines is not None:
+        cmd.append(f"-U{context_lines}")
     if base_ref:
         cmd.append(base_ref)
 

{git_commit_message-0.8.2 → git_commit_message-0.9.0}/src/git_commit_message/_gpt.py

@@ -11,20 +11,9 @@ from openai.types.responses import Response
 from os import environ
 from typing import ClassVar
 
-from tiktoken import Encoding, encoding_for_model, get_encoding
 from ._llm import LLMTextResult, LLMUsage
 
 
-def _encoding_for_model(
-    model: str,
-    /,
-) -> Encoding:
-    try:
-        return encoding_for_model(model)
-    except Exception:
-        return get_encoding("cl100k_base")
-
-
 class OpenAIResponsesProvider:
     __slots__ = (
         "_client",
@@ -50,8 +39,35 @@ class OpenAIResponsesProvider:
         model: str,
         text: str,
     ) -> int:
-        encoding = _encoding_for_model(model)
-        return len(encoding.encode(text))
+        try:
+            resp = self._client.responses.input_tokens.count(
+                model=model,
+                input=[
+                    {
+                        "role": "user",
+                        "content": [
+                            {
+                                "type": "input_text",
+                                "text": text,
+                            }
+                        ],
+                    }
+                ],
+            )
+        except Exception as exc:
+            raise RuntimeError(
+                "Token counting failed for the OpenAI provider. "
+                "Try `--chunk-tokens 0` (default) or `--chunk-tokens -1` to disable summarisation."
+            ) from exc
+
+        prompt_tokens = getattr(resp, "input_tokens", None)
+        if not isinstance(prompt_tokens, int):
+            raise RuntimeError(
+                "Token counting returned an unexpected response from the OpenAI provider. "
+                "Try `--chunk-tokens 0` (default) or `--chunk-tokens -1` to disable summarisation."
+            )
+
+        return prompt_tokens
 
     def generate_text(
         self,

{git_commit_message-0.8.2 → git_commit_message-0.9.0}/src/git_commit_message/_llamacpp.py

@@ -12,7 +12,6 @@ from typing import ClassVar, Final
 
 from openai import OpenAI
 from openai.types.chat import ChatCompletionMessageParam
-from tiktoken import Encoding, get_encoding
 
 from ._llm import LLMTextResult, LLMUsage
 
@@ -29,15 +28,6 @@ def _resolve_llamacpp_host(
     return host or environ.get("LLAMACPP_HOST") or _DEFAULT_LLAMACPP_HOST
 
 
-def _get_encoding() -> Encoding:
-    """Get a fallback encoding for token counting."""
-
-    try:
-        return get_encoding("cl100k_base")
-    except Exception:
-        return get_encoding("gpt2")
-
-
 class LlamaCppProvider:
     """llama.cpp provider implementation for the LLM protocol.
 
@@ -135,11 +125,17 @@ class LlamaCppProvider:
                 },
                 cast_to=dict,
             )
-            return response.get("total", 0)
-        except Exception:
-            # Fallback to tiktoken approximation
-            try:
-                encoding = _get_encoding()
-                return len(encoding.encode(text))
-            except Exception:
-                return len(text.split())
+        except Exception as exc:
+            raise RuntimeError(
+                "Token counting failed for the llama.cpp provider. "
+                "Try `--chunk-tokens 0` (default) or `--chunk-tokens -1` to disable summarisation."
+            ) from exc
+
+        total = response.get("total") if isinstance(response, dict) else None
+        if not isinstance(total, int):
+            raise RuntimeError(
+                "Token counting returned an unexpected response from the llama.cpp provider. "
+                "Try `--chunk-tokens 0` (default) or `--chunk-tokens -1` to disable summarisation."
+            )
+
+        return total

{git_commit_message-0.8.2 → git_commit_message-0.9.0}/src/git_commit_message/_llm.py

@@ -11,16 +11,14 @@ Provider-specific API calls live in provider modules (e.g. `_gpt.py`).
 from __future__ import annotations
 
 from babel import Locale
-from os import environ
-from typing import ClassVar, Final, Protocol
+from typing import ClassVar, Protocol
 
-
-_DEFAULT_PROVIDER: Final[str] = "openai"
-_DEFAULT_MODEL_OPENAI: Final[str] = "gpt-5-mini"
-_DEFAULT_MODEL_GOOGLE: Final[str] = "gemini-2.5-flash"
-_DEFAULT_MODEL_OLLAMA: Final[str] = "gpt-oss:20b"
-_DEFAULT_MODEL_LLAMACPP: Final[str] = "default"
-_DEFAULT_LANGUAGE: Final[str] = "en-GB"
+from ._config import (
+    resolve_language_tag,
+    resolve_model_name,
+    resolve_provider_name,
+    validate_provider_chunk_tokens,
+)
 
 
 class UnsupportedProviderError(RuntimeError):
@@ -137,49 +135,13 @@ class CommitMessageResult:
         self.total_tokens = total_tokens
 
 
-def _resolve_provider(
-    provider: str | None,
-    /,
-) -> str:
-    chosen = provider or environ.get("GIT_COMMIT_MESSAGE_PROVIDER") or _DEFAULT_PROVIDER
-    return chosen.strip().lower()
-
-
-def _resolve_model(
-    model: str | None,
-    provider_name: str,
-    /,
-) -> str:
-    if provider_name == "google":
-        default_model = _DEFAULT_MODEL_GOOGLE
-        provider_model = None
-    elif provider_name == "ollama":
-        default_model = _DEFAULT_MODEL_OLLAMA
-        provider_model = environ.get("OLLAMA_MODEL")
-    elif provider_name == "llamacpp":
-        default_model = _DEFAULT_MODEL_LLAMACPP
-        provider_model = environ.get("LLAMACPP_MODEL")
-    else:
-        default_model = _DEFAULT_MODEL_OPENAI
-        provider_model = environ.get("OPENAI_MODEL")
-
-    return model or environ.get("GIT_COMMIT_MESSAGE_MODEL") or provider_model or default_model
-
-
-def _resolve_language(
-    language: str | None,
-    /,
-) -> str:
-    return language or environ.get("GIT_COMMIT_MESSAGE_LANGUAGE") or _DEFAULT_LANGUAGE
-
-
 def get_provider(
     provider: str | None,
     /,
     *,
     host: str | None = None,
 ) -> CommitMessageProvider:
-    name = _resolve_provider(provider)
+    name = resolve_provider_name(provider)
 
     if name == "openai":
         # Local import to avoid import cycles: providers may import shared types from this module.
@@ -242,19 +204,54 @@ def _build_system_prompt(
     single_line: bool,
     subject_max: int | None,
     language: str,
+    conventional: bool = False,
     /,
 ) -> str:
     display_language: str = _language_display(language)
     max_len = subject_max or 72
     if single_line:
+        conventional_rule: str
+        if conventional:
+            conventional_rule = (
+                "Use one of these Conventional Commits subject forms: '<type>: <description>', '<type>(<scope>): <description>', '<type>!: <description>', or '<type>(<scope>)!: <description>'. "
+                "When a scope is present, it MUST be parenthesized and directly attached to the type with no spaces. "
+                "Represent breaking changes with '!' before ':' in the subject; do not output a BREAKING CHANGE footer. "
+                "Do NOT translate the Conventional prefix token ('<type>', optional '(<scope>)', optional '!'); translate only the description into the target language. "
+            )
+        else:
+            conventional_rule = (
+                "Do NOT use Conventional Commits title format. "
+                "Do not start with '<type>:' or '<type>(<scope>):' prefixes such as 'feat:', 'fix:', 'docs:', 'chore:', 'refactor:', 'test:', 'perf:', 'ci:', or 'build:'. "
+            )
         return (
             f"You are an expert Git commit message generator. "
             f"Always use '{display_language}' spelling and style. "
+            f"{conventional_rule}"
             f"Return a single-line imperative subject only (<= {max_len} chars). "
             f"Do not include a body, bullet points, or any rationale. Do not include any line breaks. "
             f"Consider the user-provided auxiliary context if present. "
             f"Return only the commit message text (no code fences or prefixes like 'Commit message:')."
         )
+
+    format_guidelines: str = ""
+    if conventional:
+        format_guidelines = (
+            "\n"
+            "- The subject line MUST use one of these forms: '<type>: <description>', '<type>(<scope>): <description>', '<type>!: <description>', or '<type>(<scope>)!: <description>'.\n"
+            "- If scope is used, it MUST be in parentheses and directly attached to type with no spaces, e.g. 'feat(parser):'.\n"
+            "- In Conventional mode, only the subject line and footer conventions are additionally constrained; keep the body structure unchanged.\n"
+            "- Keep the translated equivalent of 'Rationale:' as the final body line label; this section MUST be present.\n"
+            "- For breaking changes, use '!' immediately before ':' in the subject line.\n"
+            "- Do NOT generate any BREAKING CHANGE footer line.\n"
+            "- Do NOT translate the Conventional prefix token ('<type>', optional '(<scope>)', optional '!'). Translate only the description, bullet points, and rationale into the target language.\n"
+        )
+    else:
+        format_guidelines = (
+            "\n"
+            "- Do NOT use Conventional Commits subject prefixes.\n"
+            "- The subject MUST NOT start with '<type>:' or '<type>(<scope>):' patterns (for example: 'feat:', 'fix:', 'docs:', 'chore:', 'refactor:', 'test:', 'perf:', 'ci:', or 'build:').\n"
+        )
+
     return (
         f"You are an expert Git commit message generator. "
         f"Always use '{display_language}' spelling and style. "
@@ -282,6 +279,7 @@ def _build_system_prompt(
         f"- If few details are necessary, include at least one bullet summarising the key change.\n"
         f"- If you cannot provide any body content, still output the subject line; the subject line must never be omitted.\n"
         f"- Consider the user-provided auxiliary context if present.\n"
+        f"{format_guidelines}"
         f"Return only the commit message text in the above format (no code fences or extra labels)."
     )
 
@@ -371,14 +369,17 @@ def _build_diff_chunks(
 
         if current:
             chunks.append("".join(current))
-            current = [hunk]
-        else:
             single_tokens = provider.count_tokens(model=model, text=hunk)
             if single_tokens > chunk_tokens:
                 raise ValueError(
                     "chunk_tokens is too small to fit a single diff hunk; increase the value or disable chunking"
                 )
             current = [hunk]
+            continue
+
+        raise ValueError(
+            "chunk_tokens is too small to fit a single diff hunk; increase the value or disable chunking"
+        )
 
     if current:
         chunks.append("".join(current))
@@ -420,9 +421,10 @@ def _generate_commit_from_summaries(
     single_line: bool,
     subject_max: int | None,
     language: str,
+    conventional: bool = False,
     /,
 ) -> LLMTextResult:
-    instructions = _build_system_prompt(single_line, subject_max, language)
+    instructions = _build_system_prompt(single_line, subject_max, language, conventional)
     sections: list[str] = []
 
     if hint:
@@ -486,19 +488,26 @@ def generate_commit_message(
     chunk_tokens: int | None = 0,
     provider: str | None = None,
     host: str | None = None,
+    conventional: bool = False,
     /,
 ) -> str:
-    chosen_provider = _resolve_provider(provider)
-    chosen_model = _resolve_model(model, chosen_provider)
-    chosen_language = _resolve_language(language)
+    chosen_provider = resolve_provider_name(provider)
+    chosen_model = resolve_model_name(model, chosen_provider)
+    chosen_language = resolve_language_tag(language)
 
     llm = get_provider(chosen_provider, host=host)
 
     normalized_chunk_tokens = 0 if chunk_tokens is None else chunk_tokens
+    provider_arg_error = validate_provider_chunk_tokens(
+        chosen_provider,
+        normalized_chunk_tokens,
+    )
+    if provider_arg_error is not None:
+        raise ValueError(provider_arg_error)
 
     if normalized_chunk_tokens != -1:
         hunks = _split_diff_into_hunks(diff)
-        if normalized_chunk_tokens == 0 or normalized_chunk_tokens < 0:
+        if normalized_chunk_tokens == 0:
             chunks = ["".join(hunks) if hunks else diff]
         else:
             chunks = _build_diff_chunks(hunks, normalized_chunk_tokens, llm, chosen_model)
@@ -513,10 +522,11 @@ def generate_commit_message(
             single_line,
             subject_max,
             chosen_language,
+            conventional,
         )
         text = (final.text or "").strip()
     else:
-        instructions = _build_system_prompt(single_line, subject_max, chosen_language)
+        instructions = _build_system_prompt(single_line, subject_max, chosen_language, conventional)
         user_text = _build_combined_prompt(diff, hint)
         final = llm.generate_text(
             model=chosen_model,
@@ -541,21 +551,28 @@ def generate_commit_message_with_info(
     chunk_tokens: int | None = 0,
     provider: str | None = None,
     host: str | None = None,
+    conventional: bool = False,
     /,
 ) -> CommitMessageResult:
-    chosen_provider = _resolve_provider(provider)
-    chosen_model = _resolve_model(model, chosen_provider)
-    chosen_language = _resolve_language(language)
+    chosen_provider = resolve_provider_name(provider)
+    chosen_model = resolve_model_name(model, chosen_provider)
+    chosen_language = resolve_language_tag(language)
 
     llm = get_provider(chosen_provider, host=host)
 
     normalized_chunk_tokens = 0 if chunk_tokens is None else chunk_tokens
+    provider_arg_error = validate_provider_chunk_tokens(
+        chosen_provider,
+        normalized_chunk_tokens,
+    )
+    if provider_arg_error is not None:
+        raise ValueError(provider_arg_error)
 
     response_id: str | None = None
 
     if normalized_chunk_tokens != -1:
         hunks = _split_diff_into_hunks(diff)
-        if normalized_chunk_tokens == 0 or normalized_chunk_tokens < 0:
+        if normalized_chunk_tokens == 0:
             chunks = ["".join(hunks) if hunks else diff]
         else:
             chunks = _build_diff_chunks(hunks, normalized_chunk_tokens, llm, chosen_model)
@@ -570,6 +587,7 @@ def generate_commit_message_with_info(
             single_line,
             subject_max,
             chosen_language,
+            conventional,
         )
 
         combined_prompt = _build_combined_prompt(
@@ -586,7 +604,7 @@ def generate_commit_message_with_info(
         response_id = final_result.response_id
 
     else:
-        instructions = _build_system_prompt(single_line, subject_max, chosen_language)
+        instructions = _build_system_prompt(single_line, subject_max, chosen_language, conventional)
         combined_prompt = _build_combined_prompt(diff, hint)
 
         final_result = llm.generate_text(

{git_commit_message-0.8.2 → git_commit_message-0.9.0}/src/git_commit_message/_ollama.py

@@ -11,7 +11,6 @@ from os import environ
 from typing import ClassVar, Final
 
 from ollama import Client, ResponseError
-from tiktoken import Encoding, get_encoding
 
 from ._llm import LLMTextResult, LLMUsage
 
@@ -28,15 +27,6 @@ def _resolve_ollama_host(
     return host or environ.get("OLLAMA_HOST") or _DEFAULT_OLLAMA_HOST
 
 
-def _get_encoding() -> Encoding:
-    """Get a fallback encoding for token counting."""
-
-    try:
-        return get_encoding("cl100k_base")
-    except Exception:
-        return get_encoding("gpt2")
-
-
 class OllamaProvider:
     """Ollama provider implementation for the LLM protocol."""
 
@@ -113,10 +103,7 @@ class OllamaProvider:
         model: str,
         text: str,
     ) -> int:
-        """Approximate token count using tiktoken; fallback to whitespace split."""
-
-        try:
-            encoding = _get_encoding()
-            return len(encoding.encode(text))
-        except Exception:
-            return len(text.split())
+        raise RuntimeError(
+            "Token counting is not supported for the Ollama provider. "
+            "Try `--chunk-tokens 0` (default) or `--chunk-tokens -1` to disable summarisation."
+        )

{git_commit_message-0.8.2 → git_commit_message-0.9.0}/src/git_commit_message.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: git-commit-message
-Version: 0.8.2
+Version: 0.9.0
 Summary: Generate Git commit messages from staged changes using LLM
 Maintainer-email: Mina Her <minacle@live.com>
 License: This is free and unencumbered software released into the public domain.
@@ -47,7 +47,6 @@ Requires-Dist: babel>=2.17.0
 Requires-Dist: google-genai>=1.56.0
 Requires-Dist: ollama>=0.4.0
 Requires-Dist: openai>=2.6.1
-Requires-Dist: tiktoken>=0.12.0
 
 # git-commit-message
 
@@ -167,6 +166,18 @@ git-commit-message --one-line "optional context"
 git-commit-message --one-line --co-author 'John Doe <john.doe@example.com>'
 ```
 
+Use Conventional Commits constraints for the subject/footer only (body format is preserved):
+
+```sh
+git-commit-message --conventional
+
+# can be combined with one-line mode
+git-commit-message --conventional --one-line
+
+# co-author trailers are appended after any existing footers
+git-commit-message --conventional --co-author copilot
+```
+
 Select provider:
 
 ```sh
@@ -223,10 +234,24 @@ git-commit-message --chunk-tokens 0
 # chunk the diff into ~4000-token pieces before summarising
 git-commit-message --chunk-tokens 4000
 
+# note: for provider 'ollama', values >= 1 are not supported
+# use 0 (single summary pass) or -1 (legacy one-shot)
+git-commit-message --provider ollama --chunk-tokens 0
+
 # disable summarisation and use the legacy one-shot prompt
 git-commit-message --chunk-tokens -1
 ```
 
+Adjust unified diff context lines:
+
+```sh
+# use 5 context lines around each change hunk
+git-commit-message --diff-context 5
+
+# include only changed lines (no surrounding context)
+git-commit-message --diff-context 0
+```
+
 Select output language/locale (IETF language tag):
 
 ```sh
@@ -258,9 +283,11 @@ git-commit-message --provider llamacpp --host http://192.168.1.100:8080
 - `--provider {openai,google,ollama,llamacpp}`: provider to use (default: `openai`)
 - `--model MODEL`: model override (provider-specific; ignored for llama.cpp)
 - `--language TAG`: output language/locale (default: `en-GB`)
+- `--conventional`: apply Conventional Commits constraints to the subject and footer behavior. The body format is unchanged and still includes the translated `Rationale:` line. Breaking changes are expressed with `!` in the subject line, and `BREAKING CHANGE` footer lines are not generated.
 - `--one-line`: output subject only when no trailers are appended; with `--co-author`, output is a single-line subject plus `Co-authored-by:` trailer lines
 - `--max-length N`: max subject length (default: 72)
-- `--chunk-tokens N`: token budget per diff chunk (`0` = single summary pass, `-1` disables summarisation)
+- `--chunk-tokens N`: token budget per diff chunk (`0` = single summary pass, `-1` disables summarisation). For `ollama`, values `>= 1` are not supported.
+- `--diff-context N`: context lines in unified diff (`N >= 0`). If omitted, uses `GIT_COMMIT_MESSAGE_DIFF_CONTEXT` when set; otherwise uses Git default (usually `3`).
 - `--debug`: print request/response details
 - `--commit`: run `git commit -m <message>`
 - `--amend`: generate a message suitable for amending the previous commit (diff is from the amended commit's parent to the staged index; if nothing is staged, this effectively becomes the diff introduced by `HEAD`)
@@ -284,7 +311,8 @@ Optional:
 - `OLLAMA_HOST`: Ollama server URL (default: `http://localhost:11434`)
 - `LLAMACPP_HOST`: llama.cpp server URL (default: `http://localhost:8080`)
 - `GIT_COMMIT_MESSAGE_LANGUAGE`: default language/locale (default: `en-GB`)
-- `GIT_COMMIT_MESSAGE_CHUNK_TOKENS`: default chunk token budget (default: `0`)
+- `GIT_COMMIT_MESSAGE_CHUNK_TOKENS`: default chunk token budget (default: `0`; for `ollama`, values `>= 1` are not supported)
+- `GIT_COMMIT_MESSAGE_DIFF_CONTEXT`: default unified diff context lines (`0` or greater). If unset, Git default is used (usually `3`).
 
 Default models (if not overridden):
 

{git_commit_message-0.8.2 → git_commit_message-0.9.0}/src/git_commit_message.egg-info/SOURCES.txt

@@ -4,6 +4,7 @@ pyproject.toml
 src/git_commit_message/__init__.py
 src/git_commit_message/__main__.py
 src/git_commit_message/_cli.py
+src/git_commit_message/_config.py
 src/git_commit_message/_gemini.py
 src/git_commit_message/_git.py
 src/git_commit_message/_gpt.py

{git_commit_message-0.8.2 → git_commit_message-0.9.0}/src/git_commit_message.egg-info/requires.txt

@@ -2,4 +2,3 @@ babel>=2.17.0
 google-genai>=1.56.0
 ollama>=0.4.0
 openai>=2.6.1
-tiktoken>=0.12.0