Commits

Armin Rigo  committed 62d0e0f

This day's internal progress report and planning.

  • Participants
  • Parent commits ab98d68

Comments (0)

Files changed (2)

File sprintinfo/hildesheim2-done.txt

+
+PyPy Hildesheim2 sprint -- DONE 
+-------------------------------
+
+- do we want to continue to work with the snapshsot? does it make sense
+  during the sprint?
+  decision: during the sprint use directly the trunk, mail pypy-dev 
+  about this (holger, samuele)
+
+
+- we want to work on pypy/translator/goal/ISSUES.txt
+
+pairs::
+
+    Samuele, Carl Friedrich (progressing on External functions) 
+    DONE: refactored math and os function out of annotation/builtin and
+    rpython/rbuiltin into the extfunctable, implemented some
+    math.*, os.* functions. fixed os.stat
+
+    Open issues:
+    - DONE (for C) RPython string -> C null-terminated string 
+      conversion. The null termination char is not there!
+    
+    - DONE (__builtin__/* don't raise, the rest does)
+      The flow space kills exception handling for built-ins,
+      this is not what we want e.g. for os.* functions;
+      decide new criteria. Maybe only for thing in __builtin__
+      or list only the built-ins we want exception handling 
+      suppressed.
+
+    - DONE finish os.path.exists and os.path.isdir support
+
+     (Armin, Carl Friedrich)
+
+
+    Armin, Christian (random RTyper problems)
+    DONE: down to override:cpy_stablecompiler,
+    up to wrap/unwrap issues related to faking.
+    The hope is that we can get rid of faking.
+ 
+    DONE Completing marshal, made it into a mixed-module
+    that can be geninterped.
+    (chris wishes to redo it in interplevel)
+
+    DONE modulo some problems.
+    Inconsistency about problems on different machines.
+    PYC still untested. More tests needed.
+    Plan: enable pyc loading to bootstrap compiler
+    at app-level (Christian, Richard)
+    Todo:
+    Choose a PYC file version for PyPy (2.4.1)
+    Using this version is ok, because our compiler
+    exactly produces these. The extra functionality in
+    our marshal is not related to magic numbers.
+
+    Richard, Holger (getting rid of faked files) 
+    DONE: importing uses only os.* functions at
+    interp-level. With --file option sys.std*
+    use our own app-level implementation of files.
+    --file is on by default now when translating
+        
+    Remaining problem list:
+
+    - DONE Samuele, Holger
+      make py.py --file --nofakedmodules work
+      (crashes somehow)
+
+    - DONE (chris)
+      allways enforce geninterping of marshal, maybe
+      similar to what we did with exceptions
+
+    - DONE Samuele, Holger
+      primitive math is done. Implement applevel math,
+      have many tests (we are using compliance tests only).
+
+    - Samuele, Holger, transitional
+      unify the way stdobjspace is configured.
+      the basic objspace configuration interface
+      becomes a set of keyword arguments. The current
+      command line arguments will be mapped onto this
+      interface.
+      This also includes configuration of which modules
+      should be loaded.
+
+    - Samuele, eleumaS
+      extend the testing framework to allow to specify
+      a space with options for a test. Related to the latter.
+
+    - Holger, Armin
+      bug to be fixed: applevel test stdout is no longer
+      captured.
+
+    - Armin, and everybody else, please
+      run translation and try to avoid any faking at all.
+      An example where it still happens is EllipsisType
+      which gets wrapped by geninterplevel.
+
+    - DONE
+      Richard, Chris
+      Choose a PYC file version for PyPy (2.4.1)
+      Using this version is ok, because our compiler
+      exactly produces these. The extra functionality in
+      our marshal is not related to magic numbers.
+
+    - Richard, Chris
+      track down remaining .pyc file issues.

File sprintinfo/hildesheim2-planning.txt

 Current Main task: RTYPER issues
 -----------------------------------
 
-- do we want to continue to work with the snapshsot? does it make sense
-  during the sprint?
-  decision: during the sprint use directly the trunk, mail pypy-dev 
-  about this (holger, samuele)
+Remaining problem about pyc files::
 
+    The inconsistency problems are more or less solved now,
+    but PYC file support is disabled at the moment because
+    it is just too slow in general.
 
-- we want to work on pypy/translator/goal/ISSUES.txt
+    Bootstrapping is right but messy.  Christian will rewrite
+    marshal at interp-level at some point.
 
-pairs::
+    We won't use PYC file support for translate_pypy anyway.
+    We will construct and initialize the object space just like now,
+    and then import at app-level the 'compiler' package, and finally
+    patch the executioncontext.compiler to use this 'compiler'
+    package.  The resulting object space is what translate_pypy is
+    seeing.
 
-    Samuele, Carl Friedrich (progressing on External functions) 
-    DONE: refactored math and os function out of annotation/builtin and
-    rpython/rbuiltin into the extfunctable, implemented some
-    math.*, os.* functions. fixed os.stat
+    Ultimately the compiler and marshal should both be at
+    interp-level anyway.
 
-    Open issues:
-    - DONE (for C) RPython string -> C null-terminated string 
-      conversion. The null termination char is not there!
-    
-    - DONE (__builtin__/* don't raise, the rest does)
-      The flow space kills exception handling for built-ins,
-      this is not what we want e.g. for os.* functions;
-      decide new criteria. Maybe only for thing in __builtin__
-      or list only the built-ins we want exception handling 
-      suppressed.
+Getting rid of faking::
 
-    - DONE finish os.path.exists and os.path.isdir support
+    DONE.  no faking left if you use --file --nofakedmodules.
+    (of course in this case you can't import most modules like
+    _sre at all.)
 
-     (Armin, Carl Friedrich)
+Left::
 
-
-    Armin, Christian (random RTyper problems)
-    DONE: down to override:cpy_stablecompiler,
-    up to wrap/unwrap issues related to faking.
-    The hope is that we can get rid of faking.
- 
-    DONE Completing marshal, made it into a mixed-module
-    that can be geninterped.
-    (chris wishes to redo it in interplevel)
-
-    DONE modulo some problems.
-    Inconsistency about problems on different machines.
-    PYC still untested. More tests needed.
-    Plan: enable pyc loading to bootstrap compiler
-    at app-level (Christian, Richard)
-    Todo:
-    Choose a PYC file version for PyPy (2.4.1)
-    Using this version is ok, because our compiler
-    exactly produces these. The extra functionality in
-    our marshal is not related to magic numbers.
-
-    Richard, Holger (getting rid of faked files) 
-    DONE: importing uses only os.* functions at
-    interp-level. With --file option sys.std*
-    use our own app-level implementation of files.
-    --file is on by default now when translating
-        
-    Remaining problem list:
-
-    - DONE Samuele, Holger
-      make py.py --file --nofakedmodules work
-      (crashes somehow)
-
-    - DONE (chris)
-      allways enforce geninterping of marshal, maybe
-      similar to what we did with exceptions
-
-    - DONE Samuele, Holger
-      primitive math is done. Implement applevel math,
-      have many tests (we are using compliance tests only).
-
-    - Samuele, Holger, transitional
-      unify the way stdobjspace is configured.
+    - unify the way stdobjspace is configured.
       the basic objspace configuration interface
       becomes a set of keyword arguments. The current
       command line arguments will be mapped onto this
       This also includes configuration of which modules
       should be loaded.
 
-    - Samuele, eleumaS
-      extend the testing framework to allow to specify
+    - extend the testing framework to allow to specify
       a space with options for a test. Related to the latter.
 
-    - Holger, Armin
-      bug to be fixed: applevel test stdout is no longer
+    - bug to be fixed: applevel test stdout is no longer
       captured.
 
-    - Armin, and everybody else, please
-      run translation and try to avoid any faking at all.
-      An example where it still happens is EllipsisType
-      which gets wrapped by geninterplevel.
+    - running translate_pypy!
 
-    - DONE
-      Richard, Chris
-      Choose a PYC file version for PyPy (2.4.1)
-      Using this version is ok, because our compiler
-      exactly produces these. The extra functionality in
-      our marshal is not related to magic numbers.
+Pairing
+-------
 
-    - Richard, Chris
-      track down remaining .pyc file issues.
+configuration of the stdobjspace (holger, richard)
 
-Concerning the sync meeting of tomorrow, and given that
-we stop working this afternoon, we agreed now for
-tomorrow, that Anders should be asked to work on the 
-complicancy test failures.
+compiler + translate_pypy (christian, samuele, armin, carl friedrich)
+(trying to sub-pair internally)