The __pypy__ module

The __pypy__ module is the main entry point to special features provided by PyPy's standard interpreter. Its content depends on configuration options which may add new functionality and functions whose existence or non-existence indicates the presence of such features. These are generally used for compatibility when writing pure python modules that in CPython are written in C. Not available in CPython, and so must be used inside a if platform.python_implementation == 'PyPy' block or otherwise hidden from the CPython interpreter.

Generally available functionality

• internal_repr(obj): return the interpreter-level representation of an object.
• bytebuffer(length): return a new read-write buffer of the given length. It works like a simplified array of characters (actually, depending on the configuration the array module internally uses this).
• attach_gdb(): start a GDB at the interpreter-level (or a PDB before translation).

• newmemoryview(buffer, itemsize, format, shape=None, strides=None): create a memoryview instance with the data from buffer and the specified itemsize, format, and optional shape and strides.
• bufferable: a base class that provides a __buffer__(self, flags) method for subclasses to override. This method should return a memoryview instance of the class instance. It is called by the C-API's tp_as_buffer. bf_getbuffer.

• builtinify(func): To implement at app-level modules that are, in CPython, implemented in C: this decorator protects a function from being ever bound like a method. Useful because some tests do things like put a "built-in" function on a class and access it via the instance.

• hidden_applevel(func): Decorator that hides a function's frame from app-level

• get_hidden_tb(): Return the traceback of the current exception being handled by a frame hidden from applevel.

• lookup_special(obj, meth): Lookup up a special method on an object.

• do_what_I_mean

• resizelist_hint(...): Reallocate the underlying storage of the argument list to sizehint

• newlist_hint(...): Create a new empty list that has an underlying storage of length sizehint

• add_memory_pressure(bytes): Add memory pressure of estimate bytes. Useful when calling a C function that internally allocates a big chunk of memory. This instructs the GC to garbage collect sooner than it would otherwise.

• newdict(type): Create a normal dict with a special implementation strategy. type is a string and can be:

  • "module" - equivalent to some_module.__dict__
  • "instance" - equivalent to an instance dict with a not-changing-much set of keys
  • "kwargs" - keyword args dict equivalent of what you get from **kwargs in a function, optimized for passing around
  • "strdict" - string-key only dict. This one should be chosen automatically

• reversed_dict: Enumerate the keys in a dictionary object in reversed order. This is a __pypy__ function instead of being simply done by calling reversed(), for CPython compatibility: dictionaries are ordered in PyPY but not in Cpython2.7. You should use the collections.OrderedDict class for cases where ordering is important. That class implements __reversed__ by calling __pypy__.reversed_dict()

• dict_popitem_first: Interp-level implementation of OrderedDict.popitem(last=False).

• delitem_if_value_is Atomic equivalent to: if dict.get(key) is value: del dict[key].

  SPECIAL USE CASES ONLY! Avoid using on dicts which are specialized, e.g. to int or str keys, because it switches to the object strategy. Also, the is operation is really pointer equality, so avoid using it if value is an immutable object like int or str.

• move_to_end: Move the key in a dictionary object into the first or last position. This is used in Python 3.x to implement OrderedDict.move_to_end().

• strategy(dict or list or set): Return the underlying strategy currently used by the object

• specialized_zip_2_lists

• locals_to_fast

• set_code_callback

• save_module_content_for_future_reload

• decode_long

• side_effects_ok: For use with the reverse-debugger: this function normally returns True, but will return False if we are evaluating a debugging command like a watchpoint. You are responsible for not doing any side effect at all (including no caching) when evaluating watchpoints. This function is meant to help a bit---you can write:

  if not __pypy__.side_effects_ok():
      skip the caching logic
  

  inside getter methods or properties, to make them usable from watchpoints. Note that you need to re-run REVDB=.. pypy after changing the Python code.

• stack_almost_full: Return True if the stack is more than 15/16th full.

• pyos_inputhook: Call PyOS_InputHook() from the CPython C API

• os.real_getenv(...) gets OS environment variables skipping python code

• _pypydatetime provides base classes with correct C API interactions for the pure-python datetime stdlib module

Fast String Concatenation

Rather than in-place concatenation +=, use these to enable fast, minimal copy, string building.

• builders.StringBuilder
• builders.UnicodeBuilder

Interacting with the PyPy debug log

The following functions can be used to write your own content to the PYPYLOG.

• debug_start(category, timestamp=False): open a new section; if timestamp is True, also return the timestamp which was written to the log.
• debug_stop(category, timestamp=False): close a section opened by debug_start.
• debug_print(...): print arbitrary text to the log.
• debug_print_once(category, ...): equivalent to debug_start + debug_print + debug_stop.
• debug_flush: flush the log.
• debug_read_timestamp(): read the timestamp from the same timer used by the log.
• debug_get_timestamp_unit(): get the unit of the value returned by debug_read_timestamp().

Depending on the architecture and operating system, PyPy uses different ways to read timestamps, so the timestamps used in the log file are expressed in varying units. It is possible to know which by calling debug_get_timestamp_unit(), which can be one of the following values:

tsc

The default on x86 machines: timestamps are expressed in CPU ticks, as read by the Time Stamp Counter.

ns

Timestamps are expressed in nanoseconds.

QueryPerformanceCounter

On Windows, in case for some reason tsc is not available: timestamps are read using the win API QueryPerformanceCounter().

Unfortunately, there does not seem to be a reliable standard way for converting tsc ticks into nanoseconds, although in practice on modern CPUs it is enough to divide the ticks by the maximum nominal frequency of the CPU. For this reason, PyPy gives the raw value, and leaves the job of doing the conversion to external libraries.

Transparent Proxy Functionality

If transparent proxies are enabled (with --objspace-std-withtproxy) the following functions are put into __pypy__:

• tproxy(typ, controller): Return something that looks like it is of type typ. Its behaviour is completely controlled by the controller. See the docs about transparent proxies for detail.
• get_tproxy_controller(obj): If obj is really a transparent proxy, return its controller. Otherwise return None.

Additional Clocks for Timing

The time submodule exposes the platform-dependent clock types such as CLOCK_BOOTTIME, CLOCK_MONOTONIC, CLOCK_MONOTONIC_COARSE, CLOCK_MONOTONIC_RAW and two functions:

• clock_gettime(m) which returns the clock type time in seconds and
• clock_getres(m) which returns the clock resolution in seconds.

Extended Signal Handling

thread.signals_enbaled is a context manager to use in non-main threads.

enables receiving signals in a "with" statement. More precisely, if a signal is received by the process, then the signal handler might be called either in the main thread (as usual) or within another thread that is within a "with signals_enabled:". This other thread should be ready to handle unexpected exceptions that the signal handler might raise --- notably KeyboardInterrupt.

Integer Operations with Overflow

• intop provides a module with integer operations that have two-complement overflow behaviour instead of overflowing to longs

Functionality available on py.py (not after translation)

• isfake(obj): returns True if obj is faked.