Summary:
Changes in String to get us closer to running Rails. We are now passing
Rails specs at a 375/421 (89%) rate. String specs overall are passing at
a 572/744 (76%) rate. Most of the specs that are failing are related to
different regex implementation between .NET and Ruby. This shelveset
passed SNAP this morning.
Details:
This is a targeted set of changes that aims to implement only the String
methods that our Rails dynamic traces list. The goal is to get these
changes to maximum conformance with the spec test suite.
Overall, these are the test pass results from methods used by Rails:
Pass Total %
% 63 69 91%
-
6 6 100%
-
4 4 100%
<=> 10 11 91%
== 5 5 100%
[] 88 89 99%
chop 9 9 100%
concat 10 10 100%
delete[!] 10 11 91%
downcase[!] 7 7 100%
dump 4 4 100%
each 8 10 80%
empty? 1 1 100%
gsub[!] 26 38 68%
hash 1 1 100%
inspect 2 3 67%
length 1 1 100%
match[=~] 11 11 100%
replace 5 5 100%
scan 11 15 73%
size 1 1 100%
split 26 27 96%
strip[!] 4 5 80%
sub[!] 26 39 67%
to_i 10 12 83%
to_s 3 3 100%
to_str 3 3 100%
to_sym 2 2 100%
tr 11 11 100%
unpack 0 1 0%
upcase[!] 7 7 86%
375 421 89%
Frozen / taint related changes:
- Added Protocols.FlowTaint() helpers to add taint tracking to String
APIs - Added Protocols.AssertNotFrozen() helpers
- Simple changes in Hash and Array to clean up some of the simple frozen
/ taint specs that were failing.
Regex related changes:
- Added Match#begin, which is used by String APIs
- Fixed Match#match and Match#=~ methods so that they pass the specs
- Added Match.escape/quote and Match.last_match
- Added some constructors and an IDuplicable implementation for Regexp
builtin
Miscellaneous changes:
- 1 line change to Socket.cs that reflects changes in MutableString
- Added helper methods and a new char ctor to MutableString
- Kernel#clone, Kernel#dup, Kernel#to_s now flow taint correctly
- Fixed a bug in at_exit related to recent DLR DynamicSites refactoring
(rubycontext.cs). - Resets the baseline spec excludes to reflect new 572 passing baseline
String formatter related changes:
StringFormatter: This class contains a lot of changes to correctly
support the String#% specs. That said, right now StringFormatter really
needs a rewrite so that we align our error messages with Ruby error
messages when format strings are incorrect. Some of the changes around
how numbers are formatted will remain, but there is future work to be
done here to clean up this class. We are mostly correct in the specs
today, with just a few corner cases where we break. Some of these corner
case examples are commented out in the modulo_spec.rb file. When we
migrate to the latest Rubinius specs these will break again, which is
fine since that will be a good time for us to revisit these changes.
String related changes
MutableStringOps: The specs that we are failing today fall into one of
two camps: 1) those that are related to creating subclasses of String in
methods that return Strings.
- Added helpers to create String subclasses - due to a bug today in not
calling inherited initializers, these helpers aren’t wired up everywhere
yet today. - Added helpers for detecting frozen strings.
- Added RangeParser/IntervalParser helper classes to parse format
strings used in #delete, #count, and #tr methods. These are strings of
the form c1-c2 with negation (^) and intersection. - Started inserting ri#method documentation as comments before each
method to better capture the Ruby semantics of a method in our sources.
These docs are not as complete as the specs, which are much more
authoritative here. - Many modifications to existing methods to support flowing taint and
frozen semantics. - Fixed implementation of #<<, #concat.
- Added ‘turn-it-around’ comparison logic to String#<=>
- Added Regex support to #slice[!] methods
- Rewrote String#chomp implementation to pass 100% of specs and run on
Silverlight - Fixed implementations of #casecmp, #capitalize[!], #downcase[!],
#swapcase[!], #upcase[!], #chop[!], - Added some missing metacharacters to #dump, #inspect.
- Fixed implementation of #each, #each_line to correctly deal with
defaults - Fixed a number of bugs in #sub[!], #gsub[!]
- Removed overloads that take a string parameter. We only have overloads
that accept MutableString and object (via Protocols.CastToString()). - Cleaned up implementation of #index
- Added new implementations of #delete[!] and #count
- Added #=~ and cleaned up #match
- Added #scan implementation
- Cleaned up #succ[!]
- Added full implementation of #split[!]
- Added an implementation of #to_i that uses Ruby tokenizer to generate
output - Added full implementation of #replace
- Added full implementation of #tr[!]
There are also a ton of other DLR and Ruby-related changes in this
changeset, but these are the most visible ones at this time. Will
summarize details in a future mail.
Thanks,
-John