Overview
=========
-Although focus has and still is on graphical APIs, apitrace has an
-infrastructure to trace generic APIs:
+Although focus was and still is on graphical APIs, apitrace has a
+generic infrastructure to trace any kind of API:
- * The APIs types and calls are specified in Python files in spec
+ * the APIs types and calls are specified in Python files in specs
+ sub-directory;
- * there is a type hierarchy in specs\stdapi.py , capable of representing
+ * there is a type hierarchy in specs/stdapi.py, capable of representing
most types in C language, and additional semantic metadata
- * Python scripts generate C code to trace and serialize calls to disk, and
- vice-versa.
+ * Python scripts generate C++ code to trace and serialize calls parameters to
+ a file, and vice-versa.
* Visitor software design pattern is used to navigate over the types.
overriden by derived classes, allowing to easily handle cases that need
special treatment without sacrifycing code reuse.
-There are several main layers in apitrace. Too many to show in a single graph,
-so below only those relevant for GL are shown:
+apitrace's architecture is composed of several layers. Too many to show in a
+single graph, so only those relevant for OpenGL API are shown below:
specs
^
/ | \
glxtrace wgltrace cgltrace
-And here is a quick synopsis of what the layers do:
+Here is a quick synopsis of what the layers do:
* specs -- specification of the API, expressed in a Python class hierarchy
* dispatch -- runtime dispatch of calls to DLLs (open the DLL, get the symbol
address, and call it passing all arguments as-is)
- * helpers -- helper functions to determine sizes of array/blob/etc, and other.
- It often needs to dispatch calls to give the answers.
+ * helpers -- helper functions to determine sizes of arrays, blobs, etc. It
+ often needs to dispatch calls to give the answers.
* trace -- generate C++ code for tracing an API based on its spec
Coding Style
============
-These are guidelines for new code. Some of existing hasn't been updated to
-these conventions yet.
+These are guidelines for new code. Admittedly some of the existing code hasn't
+been updated to follow these conventions yet.
Whitespace (all languages):
Feature development:
* Existing features in master branch should not degrade at any time, for any
- platform. (Unless it is not widely used and there is agreement.)
+ platform. (Unless they are seldom used or redundant and there is agreement.)
-* It's fine to add new features for only some platforms.
+ * In particular, new features / changes must not introduce any sort of
+ instability when tracing.
-* Non-trivial changes should be staged in a branch, to enable peer-review and
- regression testing. Branch should be deleted once code has been merged.
+ While application developers and driver developers may be able to
+ workaround quirks in apitrace, we want to be able to obtain traces from
+ non-technical end-users with minimal intervention.
+
+ This implies that tracing should not make any non-standard assumptions, and
+ care must be taken to ensure the tracing code is robust against invalid
+ parameters, multiple threads, etc.
+
+* It's fine to add new features for only some platforms or APIs.
+
+* Non-trivial changes should be staged in a branch, to allow review and
+ regression testing. Feature branches should be deleted once they have been
+ merged.
* Releases are tagged commits from master. There are no stable branches.
* No backwards compatibility guarantees for derived data (ASCII dumps, state,
images, etc).
-* There should be no gratuitous change to command line tool interfaces, but no
+* There should be no gratuitous changes to command line tool interfaces, but no
guarantees are given.
projects / apitrace / blobdiff