Add interpreter constraints option and use constraints to search for compatible interpreters at exec time #427

CMLivingston · 2017-10-20T19:36:49Z

Problem:
PEX does not support interpreter compatibility ranges for selecting an interpreter to build/run a PEX with. Twitter is aiming at killing off PEX_PYTHON overriding and replacing with a PEX_PYTHON_PATH in /etc/pexrc, which will contain a path to directories containing canonical interpreters for searching against at pex bootstrap re-exec time. Interpreter constraints on a PEX should inform the searching of PEX_PYTHON_PATH (if present in the env) and error out if no compatible interpreter can be found, replacing the searching of $PATH as a fallback.

Solution:
Add an interpreter constraints option to pex CLI and write these constraints to PEX-INFO metadata. Add an env variable to variables.py to support the setting of PEX_PYTHON_PATH. When the bootstrapper pex re-execs at runtime, read constraints from PEX-INFO and attempt to resolve an interpreter to exec with from the PPP. If no interpreter can be found in PPP, error out indicating that no match can be found.

Result:
PEXes will be able to be built and executed with specified interpreter constraints. Users will be able to define and control interpreter selection in managed development environments by setting the PEX_PYTHON_PATH so that PEXes are executed with supported interpreters only.

…NFO; filter interpreters used to build/run the pex based on these constraints.

…d add a unit test for new method

…raints matching and search api for pex bootstrap re-exec; add associated unit + integration tests

CMLivingston · 2017-10-23T23:13:26Z

pex/interpreter_constraints.py

@@ -0,0 +1,58 @@
+# Copyright 2014 Pants project contributors (see CONTRIBUTORS.md).


This new library is extent of the intersecting interpreter constraints logic that makes sense to move out of Pants and into Pex. It will be cleanly consumable by Pants and the only modification I made is to add a meet_all_constraints kwarg for meeting all filters passed instead of at least one, which is backward compatible with the Pants use cases.

I operated on the assumption that the other methods of this lib have been tested or are functioning properly so I did not add any new tests for them but I am more than happy to.

Should this get bumped to 2017?

…ific installations

…eter selection tests green; choose highest compatible interpreter in pex python path instead of lowest.

…ston/pex into clivingston/add-pex-python-path

CMLivingston · 2017-11-01T19:21:45Z

pex/interpreter_constraints.py

+from .interpreter import PythonIdentity
+
+
+def _matches(interpreter, filters, meet_all_constraints=False):


I added meet_all_constraints because for some reason the Pants usage of this code (currently in Pants codebase) only requires a match against one of the filters to be considered a "matching interpreter".

…le, but maintain support for pex_python if still set. Add relevant integration tests to test each of the test cases that deal with interpreter constraints.

kwlzn

initial pass through the sources - will circle back to review tests.

kwlzn · 2017-11-15T01:16:57Z

pex/bin/pex.py

+  if options.interpreter_constraint:
+    constraints = options.interpreter_constraint
+    # TODO: add check to see if constraints are mutually exclusive (bad) so no time is wasted
+    check_requirements_are_well_formed(constraints)


how about calling this validate_constraints() and moving the TODO here into it's function definition?

kwlzn · 2017-11-15T01:44:08Z

pex/bin/pex.py

+    # TODO: add check to see if constraints are mutually exclusive (bad) so no time is wasted
+    check_requirements_are_well_formed(constraints)
+    interpreters = list(matched_interpreters(interpreters, constraints, meet_all_constraints=True))
+


seems to me like in the case where no --python args are explicitly passed, that interpreters here pre-constraint filtration will only contain the currently executing interpreter which is probably insufficient for any users targeting e.g. python3 with constraints while executing pex with python2.

this would mean that anytime a user passes --constraint they'd also need to pass one or more --python args to surface the interpreters to actually apply the constraints to - which feels awkward.

I think we'll probably need to:

make --python mutually exclusive with --constraints - or at least warn when both are passed.

fall back to path or PPP search when --constraints is passed without --python.

potentially deprecate --python in favor of --constraints, if it makes sense.

this will pave a way for a future where both build and runtime interpreter selection behave identically, afaict.

I like the idea of a combination of 1 and 2. Disallow both options as using both seems like an anti-pattern and fall back to PPP/PATH when --constraints is passed without --python.

kwlzn · 2017-11-15T01:46:08Z

pex/bin/pex.py

@@ -535,7 +545,8 @@ def build_pex(args, options, resolver_option_builder):
    # options.preamble_file is None
    preamble = None

-  interpreter = _lowest_version_interpreter(interpreters)
+  # TODO: make whether to take min or max version interpreter configurable
+  interpreter = min(interpreters)


this TODO might be better as a github issue link instead?

kwlzn · 2017-11-15T02:07:53Z

pex/interpreter_constraints.py

+  if meet_all_constraints:
+    return all(interpreter.identity.matches(filt) for filt in filters)
+  else:
+    return any(interpreter.identity.matches(filt) for filt in filters)


how about parameterizing the function applied here to enforce DRY?

check = all if meet_all_constraints else any return check(interpreter.identity.matches(filt) for filt in filters)

kwlzn · 2017-11-15T02:13:41Z

pex/interpreter_constraints.py

+    try:
+      PythonIdentity.parse_requirement(req)
+    except ValueError as e:
+      from .common import die


this import should move to the top of the file

kwlzn · 2017-11-15T03:28:15Z

pex/testing.py

+
+def bootstrap_python_installer():
+  install_location = os.getcwd() + '/.pyenv_test'
+  if not os.path.exists(install_location) or not os.path.exists(install_location + '/bin/pyenv'):


kwlzn · 2017-11-15T03:29:10Z

pex/testing.py

+  install_location = os.getcwd() + '/.pyenv_test/versions/' + version
+  if not os.path.exists(install_location):
+    os.environ['PYENV_ROOT'] = os.getcwd() + '/.pyenv_test'
+    subprocess.call([os.getcwd() + '/.pyenv_test/bin/pyenv', 'install', version])


os.path.join throughout this function.

kwlzn · 2017-11-15T03:29:56Z

pex/variables.py

+  def PEX_PYTHON_PATH(self):
+    """String
+
+    A colon-seperated string containing paths to binaires of blessed Python interpreters


kwlzn · 2017-11-15T03:33:04Z

pex/variables.py

+    of a second execution of maybe_reexec_pex if neither PEX_PYTHON or PEX_PYTHON_PATH are set but
+    interpreter constraints are specified.
+    """
+    value = self._environ.get('SHOULD_EXIT_REEXEC', 'False')


would expect the environ var to map 1:1 to the property name here.

kwlzn · 2017-11-15T03:34:21Z

pex/variables.py

+    interpreter constraints are specified.
+    """
+    value = self._environ.get('SHOULD_EXIT_REEXEC', 'False')
+    return False if value == 'False' else True


would be better to not rely on the string value here as anything other than indicator of truthyness, i.e.:

return bool(self._environ.get(..., False))

kwlzn · 2017-11-15T03:36:07Z

noting that this also seems to be failing on the pypy shard after a few restarts.

UnrememberMe

looks good overall, a couple of cleanup suggestions

UnrememberMe · 2017-11-17T19:01:43Z

pex/pex_bootstrapper.py

+  """Find all compatible interpreters on the system within the supplied constraints and use
+     PEX_PYTHON_PATH env variable if it is set. If not, fall back to interpreters on $PATH.
+  """
+  if pex_python_path:


same codes as pex/bin/pex.py? If so, we should extract this to a library.

UnrememberMe · 2017-11-17T19:01:50Z

pex/interpreter_constraints.py

+  :param meet_all_constraints: whether to match against all filters.
+    Defaults to matching interpreters that match at least one filter.
+  """
+  for match in _matching(interpreters, constraints, meet_all_constraints):


Seems like we can simplify to

def matched_interpreters(interpreters, constraints, meet_all_constraints=False): _matching(interpreters, constraints, meet_all_constraints=meet_all_constraints)

you'd need to return the generator, but it'd probably be simpler to just fold _matching() directly into this method.

…lection

…x bootstrapper to be cleaner and simpler

kwlzn · 2017-11-23T01:20:19Z

pex/variables.py

+    """Read pex runtime configuration variables from a pexrc file.
+
+       :param rc: an absolute path to a pexrc file
+       :return ret_vars: a dict object containing key values pairs found in processed pexrc files


these and docstring continuations should indent to align with the opening """, e.g.

"""The docstring summary which might spill onto a new line. Extended summary would go here. :param etc: Would go here. """

…ular testing for the interpreter selection method in pex bootstrapper

CMLivingston · 2017-11-28T01:14:20Z

pex/variables.py

+                    os.path.join(os.path.dirname(sys.argv[0]), '.pexrc'),
+                    os.getcwd()]
+    if not rc:
+      rc_locations.append('~/.pexrc')


The only time ~/.pexrc is the default is if we are calling the Variables constructor; the from_rc api requires a string so I added support for adding ~/.pexrc if the input is blank. I suppose I could make it a default arg or pass it in PEX CLI...

Added it as default arg.

seems to me like ~/.pexrc needs to be considered in all cases (i.e. as a static member of rc_locations) or else we aren't reading from all of the places that pexrc configuration can exist. (essentially, I don't see any existing supporting usage of Variables in pex/pants to warrant keeping the rc param here the way it is).

kwlzn

looking great - handful of remaining feedback and a larger question on whether PEX_PYTHON should still be supported at all.

kwlzn · 2017-11-29T00:55:00Z

pex/bin/pex.py

@@ -525,6 +530,16 @@ def build_pex(args, options, resolver_option_builder):
      for interpreter in options.python or [None]
    ]

+  if options.interpreter_constraint:
+    # Overwrite the current interpreter as defined by sys.executable.


this comment doesn't make sense here.

kwlzn · 2017-11-29T01:41:11Z

pex/interpreter_constraints.py

+def _matching(interpreters, constraints, meet_all_constraints=False):
+  for interpreter in interpreters:
+    check = all if meet_all_constraints else any
+    if check(interpreter.identity.matches(filt) for filt in constraints):


check should be defined outside of the for body

kwlzn · 2017-11-29T01:41:40Z

pex/interpreter_constraints.py

+      PythonIdentity.parse_requirement(req)
+    except ValueError as e:
+      die("Compatibility requirements are not formatted properly: %s" % str(e))
+


specify the requirement that failed to parse's string.

kwlzn · 2017-11-29T02:29:39Z

pex/pex_bootstrapper.py

  if not target:
    die('Failed to find interpreter specified by PEX_PYTHON: %s' % target)
+  elif compatibility_constraints:


s/elif/if - these all act as guard clauses.

the e string contains this requirement.

_matching got folded into the matched_interpreters method so I will remove the method.

kwlzn · 2017-11-29T02:37:57Z

pex/pex_bootstrapper.py

+  with TRACER.timed('Selecting runtime interpreter based on pexrc', V=3):
+    if ENV.PEX_PYTHON and not ENV.PEX_PYTHON_PATH:
+      # preserve PEX_PYTHON re-exec for backwards compatibility
+      selected_interpreter = _select_pex_python_interpreter(ENV.PEX_PYTHON,


I know we discussed this earlier - but what was the ultimate reasoning here to leave PEX_PYTHON functionality as a dual path thing alongside PPP?

I was sort of thinking we could yank PEX_PYTHON in favor of PPP and cut a new minor version of pex w/ doc'd release notes as a cutover. it was the configuration (e.g. /etc/pexrc) that we'd need to keep there throughout the lifetime of old pex runtimes - but afaict, we can yank this in pex itself and simplify things. thoughts?

I agree about the simplification. We discussed leaving it in as a way to give us flexibility when migrating the /etc/pexrc configs in prod and in MDE, as well as provide backwards interoperability for pex users who want to continue using a single python interpreter. However, IIRC, I think the main reason was to give us flexibility with the whole migration of /etc/pexrc and to give us a safety net in case we find that PEX_PYTHON_PATH ends up disastrously breaking things or causing other unexpected problems.

I like the idea of a safety net and backwards compatibility (at least for now), and I think it wouldn't be a bad idea to keep it in for now until we have success using PPP internally, maybe with a reasonable subset of Python users successfully running and deploying py3 PEX both locally and in prod? What do you think? @UnrememberMe any thoughts here?

works for me. in that case, would be good to link to a ticket here that covers the eventual removal of PEX_PYTHON.

kwlzn · 2017-11-29T03:38:31Z

pex/pex_builder.py

+    if interpreter:
+      self._pex_info = pex_info or PexInfo.default(interpreter)
+    else:
+      self._pex_info = pex_info or PexInfo.default()


this could be simplified to:

self._pex_info = pex_info or PexInfo.default(interpreter)

given interpreter is None as the default here.

kwlzn · 2017-11-29T04:05:26Z

pex/variables.py

+                    os.path.join(os.path.dirname(sys.argv[0]), '.pexrc'),
+                    os.getcwd()]
+    if not rc:
+      rc_locations.append('~/.pexrc')


seems to me like ~/.pexrc needs to be considered in all cases (i.e. as a static member of rc_locations) or else we aren't reading from all of the places that pexrc configuration can exist. (essentially, I don't see any existing supporting usage of Variables in pex/pants to warrant keeping the rc param here the way it is).

kwlzn · 2017-11-29T04:09:36Z

pex/variables.py

+    rc_locations = ['/etc/pexrc',
+                    rc,
+                    os.path.join(os.path.dirname(sys.argv[0]), '.pexrc'),
+                    os.getcwd()]


afaict, this should reference an rcfile, not the current working dir directly?

but maybe it's not actually needed?

I think it is needed for the case where the pex you are invoking and the .pexrc is in the cwd.
Also, idk why the pi = interpreter or PythonInterpreter.get() line was not already changed to that, I am not sure why this reverted, but I will re-add, sorry about that.

kwlzn · 2017-11-29T04:15:57Z

pex/bin/pex.py

+    # affect usages of the interpreter(s) specified by the "--python" command line flag.
+    constraints = options.interpreter_constraint
+    validate_constraints(constraints)
+    rc_variables = Variables.from_rc(build_time_rc_dir=os.path.dirname(options.pex_name))


its still puzzling to me why we need to look at a .pexrc relative to the output pex name for PPP configuration at build time.. can you clarify that with a comment here?

if this is purely for tests, we should probably re-evaluate.

The use case is when the output pex is not in cwd, i.e. -o dir/output.pex. In that case, we should be supporting the case where the .pexrc is in dir as per a precedent set in the tests, however I see this being practical beyond testing situations.

Previously, the check was for os.path.dirname(sys.argv[0]) to find this .pexrc in the same dir as the output pex at runtime, which works fine because sys.argv[0] is the pex itself, however, at build time, sys.argv[0] is the pex cli binary, which can be different than the dir of the output pex i.e. pex requests -o dir/output.pex. In this case, where I have .pexrc in dir, special accommodations are needed to make sure that the pex build routine can find that pexrc at build time.

Like you say, this precedent is set largely by the tests, however having a pexrc in the same location as the pex itself seems like a reasonable case to cover to me.

kwlzn · 2017-11-29T04:22:11Z

pex/variables.py

+  def PEX_PYTHON_PATH(self):
+    """String
+
+    A colon-separated string containing binaires of blessed Python interpreters


s/binaires/paths/

I will remove rc.

…Cleanup docs and refactor a few one-liners.

UnrememberMe · 2017-12-01T00:05:37Z

pex/bin/pex.py

@@ -289,6 +291,25 @@ def configure_clp_pex_environment(parser):
           'can be passed multiple times to create a multi-interpreter compatible pex. '
           'Default: Use current interpreter.')

+  group.add_option(
+    '--interpreter-constraint',
+    dest='interpreter_constraint',


IIRC, a-b-c will be automatically convert to a_b_c for variable names, so you don't necessarily need to specify dest here.

UnrememberMe · 2017-12-01T00:06:14Z

pex/bin/pex.py

+         'times.')
+
+  group.add_option(
+    '--rcfile',


suggestion: for consistency, change option to --rc-file

UnrememberMe · 2017-12-01T00:09:51Z

pex/interpreter_constraints.py

+      PythonIdentity.parse_requirement(req)
+    except ValueError as e:
+      die("Compatibility requirements are not formatted properly: %s" % str(e))
+


second. Including the failed string will help user to debug.

UnrememberMe · 2017-12-01T00:24:35Z

pex/pex_bootstrapper.py

+               'PEX_PYTHON_PATH="%s", COMPATIBILITY_CONSTRAINTS="%s"'
+               % (cmdline, sys.executable, ENV.PEX_PYTHON, ENV.PEX_PYTHON_PATH,
+                  compatibility_constraints))
+    os.execve(selected_interpreter, cmdline, ENV.copy())


won't this become MyPython MyPython original_params since on line 152, cmdline has already prepended selected_interpreter?

the first arg of cmdline is ignored in the os.execve API (not sure why exactly, but it is).

UnrememberMe

Thanks for the answer. Ship it!

lgvital · 2017-12-01T05:00:07Z

pex/interpreter_constraints.py

+
+
+def validate_constraints(constraints):
+  # TODO: add check to see if constraints are mutually exclusive (bad) so no time is wasted


Did this TODO exist already when you moved it? If not, can we start an issue for it to track?

lgvital · 2017-12-01T05:01:27Z

pex/interpreter_constraints.py

+    pex, using the Requirement-style format, e.g. ``'CPython>=3', or just ['>=2.7','<3']``
+    for requirements agnostic to interpreter class.
+  :param meet_all_constraints: whether to match against all filters.
+    Defaults to matching interpreters that match at least one filter.


Should we add a line to this docstring for what exactly we're yielding?

lgvital · 2017-12-01T05:02:11Z

pex/pex_bootstrapper.py

+  with TRACER.timed('Selecting runtime interpreter based on pexrc', V=3):
+    if ENV.PEX_PYTHON and not ENV.PEX_PYTHON_PATH:
+      # preserve PEX_PYTHON re-exec for backwards compatibility
+      # TODO: Kill this off completely in favor of PEX_PYTHON_PATH


Start an issue for this TODO?

It is on the line below this one.

lgvital · 2017-12-01T05:03:17Z

pex/pex_info.py

@@ -123,6 +123,8 @@ def __init__(self, info=None):
                       '%s of type %s' % (info, type(info)))
    self._pex_info = info or {}
    self._distributions = self._pex_info.get('distributions', {})
+    # cast as set because pex info from json must store interpreter_constraints as a list


Not sure I understand this comment, or maybe it's a typo. You're casting it as a set, but interpreter constraints needs to be stored as a list?

Reading in a pex info using from_pex or from_json will instantiate a PexInfo object with json loaded as a dict. Interpreter constraints get serialized to json as a list b.c. json does not support Python sets, so casting it here is to cover the case where we read in a populated pex-info from json (which will be a list), however we want to store it internally as a set.

lgvital

Looks good, just some minor nits and questions.

kwlzn

lgtm, one last round of minor stuff.

kwlzn · 2017-11-29T23:13:10Z

pex/pex_bootstrapper.py

+  with TRACER.timed('Selecting runtime interpreter based on pexrc', V=3):
+    if ENV.PEX_PYTHON and not ENV.PEX_PYTHON_PATH:
+      # preserve PEX_PYTHON re-exec for backwards compatibility
+      selected_interpreter = _select_pex_python_interpreter(ENV.PEX_PYTHON,


works for me. in that case, would be good to link to a ticket here that covers the eventual removal of PEX_PYTHON.

kwlzn · 2017-11-30T23:57:36Z

pex/bin/pex.py

+    default=None,
+    help='A path to a pexrc file. NOTE: this flag is for testing purposes only. It is to '
+         'be used in the case that a pexrc lives in the same directory as the created '
+         'output pex as is the case with the pex integration tests.')


how about shortening this to:

An additional path to a pexrc file to read during configuration parsing. Used primarily for testing.

kwlzn · 2017-12-01T00:27:29Z

pex/bin/pex.py

+    pexrc = None
+    if options.rc_file:
+      pexrc = options.rc_file
+    rc_variables = Variables.from_rc(rc=pexrc)


given options.rc_file defaults to None, can reduce this to:

rc_variables = Variables.from_rc(rc=options.rc_file)

kwlzn · 2017-12-01T22:24:30Z

pex/pex_bootstrapper.py

+      try:
+        interpreters.append(PythonInterpreter.from_binary(binary))
+      except Executor.ExecutionError:
+        TRACER.log("Python interpreter %s in PEX_PYTHON_PATH failed to load properly." % binary)


this should probably get logged as a hard warning (i.e. print(msg, file=sys.stderr)) vs masked in tracer logging only.

kwlzn · 2017-12-01T23:09:49Z

pex/variables.py

+    """Read pex runtime configuration variables from a pexrc file.
+
+    :param rc: an absolute path to a pexrc file.
+    :return ret_vars: a dict object containing key values pairs found in processed pexrc files


specifying an arg name for a return doesn't make sense.

should look like:

:return: A dict of key value pairs found in processed pexrc files. :rtype: dict

kwlzn · 2017-12-01T23:10:49Z

pex/variables.py

+    rc_locations = ['/etc/pexrc',
+                    '~/.pexrc',
+                    os.path.join(os.path.dirname(sys.argv[0]), '.pexrc'),
+                    os.path.join(os.getcwd(), '.pexrc')]


did the cwd variant here actually end up being needed given the new --rcfile flag? if so, why?

seems to me like tests were passing without it before?

kwlzn · 2017-12-01T23:21:03Z

tests/test_integration.py

+  with temporary_dir() as td:
+    pexrc_path = os.path.join(td, '.pexrc')
+    with open(pexrc_path, 'w') as pexrc:
+      pex_python = os.getcwd() + '/.pyenv_test/versions/3.6.3/bin/python3.6'


this should get the path from ensure_python_interpreter vs hardcoding.

kwlzn · 2017-12-01T23:21:10Z

tests/test_integration.py

+    # test PEX_PYTHON with incompatible constraints
+    pexrc_path = os.path.join(td, '.pexrc')
+    with open(pexrc_path, 'w') as pexrc:
+      pex_python = os.getcwd() + '/.pyenv_test/versions/2.7.10/bin/python2.7'


this should get the path from ensure_python_interpreter vs hardcoding.

kwlzn · 2017-12-01T23:23:54Z

tests/test_variables.py

@@ -84,27 +85,30 @@ def test_pex_get_kv():


 def test_pex_from_rc():
-  with named_temporary_file(mode='w') as pexrc:
+  with open('.pexrc', 'w') as pexrc:


I think the named_temporary_file before was slightly safer?

Fix failing tests in master that have gone undetected (as a result of bad PATH setting in #427).

Chris Livingston added 4 commits October 20, 2017 12:12

Add interpreter constraints option to pex CLI and write data to PEX-I…

460a9ae

…NFO; filter interpreters used to build/run the pex based on these constraints.

Improve integration tests

45e4640

Move helper function to interpreter constraints lib, refactor lib, an…

7ae3b61

…d add a unit test for new method

Add pex python path to supported env variables; add interpreter const…

a6175dd

…raints matching and search api for pex bootstrap re-exec; add associated unit + integration tests

CMLivingston commented Oct 23, 2017

View reviewed changes

CMLivingston changed the title ~~[WIP] Add interpreter constraints option and use constraints to search for compatible interpreters at exec time~~ Add interpreter constraints option and use constraints to search for compatible interpreters at exec time Oct 23, 2017

Chris Livingston added 12 commits October 23, 2017 17:51

Resolve interpreters for testing from .tox dir instead of system-spec…

528d93e

…ific installations

Refactor tests that use Python interpreters

4fc655d

Add mock patching for pex python path search api test

ca8a9c6

Add interpreter bootstrapping as test helper method, make all interpr…

02fa8c5

…eter selection tests green; choose highest compatible interpreter in pex python path instead of lowest.

Travis CI debugging

64c2b0c

Add pytest.mark modification to skip test in CI

3408fb3

Travis CI debug

a1c4fed

Travis CI debug

cef5bf5

Merge branch 'clivingston/add-pex-python-path' of github.com:CMLiving…

2957b80

…ston/pex into clivingston/add-pex-python-path

Travis CI debug 2

0877ad0

Travis CI test 3

762e9d3

Remove logging

ebf830a

CMLivingston commented Nov 1, 2017

View reviewed changes

Modify pex bootstrapper to repsect pex_python_path environment variab…

233918e

…le, but maintain support for pex_python if still set. Add relevant integration tests to test each of the test cases that deal with interpreter constraints.

CMLivingston force-pushed the clivingston/add-pex-python-path branch from 5c435f1 to 233918e Compare November 8, 2017 20:11

kwlzn reviewed Nov 15, 2017

View reviewed changes

Chris Livingston added 2 commits November 15, 2017 18:35

Add validation for conflicting constraints

c0ef007

Refactor and clean up code to enfore DRY

0ffe8c2

CMLivingston force-pushed the clivingston/add-pex-python-path branch 2 times, most recently from 5877b3c to 851cc53 Compare November 17, 2017 18:35

UnrememberMe reviewed Nov 17, 2017

View reviewed changes

Enforce interpreter selection at build time in addition to runtime se…

b5a86ea

…lection

CMLivingston force-pushed the clivingston/add-pex-python-path branch from 96af91b to b5a86ea Compare November 20, 2017 21:42

Chris Livingston added 3 commits November 22, 2017 15:45

Refactor methods and clean up doc strings/literals

3a143d1

Refactor testing methods for readability, refactor TRACER logic in pe…

b045ba5

…x bootstrapper to be cleaner and simpler

Refactor tests to improve readability

66ef8ee

kwlzn reviewed Nov 23, 2017

View reviewed changes

Move build time pex rc reading logic into Variables and add more gran…

6a2bd58

…ular testing for the interpreter selection method in pex bootstrapper

CMLivingston commented Nov 28, 2017

View reviewed changes

Add default arg to from_rc

8311619

kwlzn reviewed Nov 29, 2017

View reviewed changes

Chris Livingston added 3 commits November 29, 2017 13:18

Remove rc param from Variables constructor and modify related tests. …

e7590c2

…Cleanup docs and refactor a few one-liners.

Add a --rcfile flag to allow pexrc usage in integration tests

6b3e551

Add comment regarding PEX_PYTHON

34773e5

UnrememberMe reviewed Dec 1, 2017

View reviewed changes

UnrememberMe approved these changes Dec 1, 2017

View reviewed changes

lgvital reviewed Dec 1, 2017

View reviewed changes

Add comments to interpreter constraints and cleanup docstring

508226f

kwlzn approved these changes Dec 1, 2017

View reviewed changes

Final cleanups to docs and minor refactoring for tests

524456b

kwlzn approved these changes Dec 2, 2017

View reviewed changes

kwlzn merged commit df326ff into pex-tool:master Dec 2, 2017

This was referenced Dec 4, 2017

Release 1.2.14 #433

Closed

pex 1.2.14 breaks entrypoint targeting when PEX_PYTHON is present #434

Closed

kwlzn mentioned this pull request May 11, 2018

master, travis are silently broken #461

Closed

CMLivingston mentioned this pull request May 11, 2018

Fix failing tests in master #466

Merged

kwlzn pushed a commit that referenced this pull request May 12, 2018

Fix failing tests in master (#466)

4317c14

Fix failing tests in master that have gone undetected (as a result of bad PATH setting in #427).

jsirois mentioned this pull request Feb 15, 2019

Consider killing --rcfile option. #674

Open

jsirois mentioned this pull request Mar 9, 2020

Support arbitrary interpreter constraint expressions. #918

Closed

		@@ -0,0 +1,58 @@
		# Copyright 2014 Pants project contributors (see CONTRIBUTORS.md).

		from .interpreter import PythonIdentity


		def _matches(interpreter, filters, meet_all_constraints=False):



		def validate_constraints(constraints):
		# TODO: add check to see if constraints are mutually exclusive (bad) so no time is wasted

Add interpreter constraints option and use constraints to search for compatible interpreters at exec time #427

Add interpreter constraints option and use constraints to search for compatible interpreters at exec time #427

Conversation

CMLivingston commented Oct 20, 2017

CMLivingston Oct 23, 2017 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

kwlzn left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

kwlzn commented Nov 15, 2017

UnrememberMe left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

CMLivingston Nov 28, 2017 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

kwlzn left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

kwlzn Nov 29, 2017 • edited Loading

Choose a reason for hiding this comment

kwlzn Nov 29, 2017 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

CMLivingston Nov 29, 2017 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

UnrememberMe left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

lgvital left a comment

Choose a reason for hiding this comment

kwlzn left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

kwlzn Dec 1, 2017 • edited Loading

CMLivingston Oct 23, 2017 •

edited

Loading

CMLivingston Nov 28, 2017 •

edited

Loading

kwlzn Nov 29, 2017 •

edited

Loading

kwlzn Nov 29, 2017 •

edited

Loading

CMLivingston Nov 29, 2017 •

edited

Loading

kwlzn Dec 1, 2017 •

edited

Loading