Roll Observatory packages and add a roll script

packages/intl/CHANGELOG.md

Index: packages/intl/CHANGELOG.md
diff --git a/packages/intl/CHANGELOG.md b/packages/intl/CHANGELOG.md
new file mode 100644
index 0000000000000000000000000000000000000000..0469e49b108a50d244b722331b88b680e9e1c598
--- /dev/null
+++ b/packages/intl/CHANGELOG.md
@@ -0,0 +1,159 @@
+## 0.12.4+2
+  * update analyzer to '<0.27.0'
+
+## 0.12.4+1
+  * Allow the name of an Intl.message to be "ClassName_methodName", as
+    well as "functionName". This makes it easier to disambiguate
+    messages with the same name but in different classes.
+
+## 0.12.4
+  * Handle spaces in ARB files where we didn't handle them before, and
+  where Google translation toolkit is now putting them.
+
+## 0.12.3
+
+  * Use latest version of 'analyzer' and 'args' packages.
+
+## 0.12.2+1
+  * Adds a special locale name "fallback" in verifiedLocale. So if a translation
+  is provided for that locale and has been initialized, anything that doesn't
+  find a closer match will use that locale. This can be used instead of having
+  it default to the text in the original source messages.
+
+## 0.12.1
+  * Adds a DateFormat.parseLoose that accepts mixed case and missing
+  delimiters when parsing dates. It also allows arbitrary amounts of
+  whitespace anywhere that whitespace is expected. So, for example,
+  in en-US locale a yMMMd format would accept "SEP 3   2014", even
+  though it would generate "Sep 3, 2014". This is fairly limited, and
+  its reliability in other locales is not known.
+
+## 0.12.0+3
+  * Update pubspec dependencies to allow analyzer version 23.
+
+## 0.12.0+2
+  * No user impacting changes. Tighten up a couple method signatures to specify
+  that int is required.
+
+## 0.12.0+1
+  * Fixes bug with printing a percent or permille format with no fraction
+  part and a number with no integer part. For example, print 0.12 with a
+  format pattern of "#%". The test for whether
+  there was a printable integer part tested the basic number, so it ignored the
+  integer digits. This was introduced in 0.11.2 when we stopped multiplying
+  the input number in the percent/permille case.
+
+## 0.12.0
+  * Make withLocale and defaultLocale use a zone, so async operations
+    inside withLocale also get the correct locale. Bumping the version
+    as this might be considered breaking, or at least
+    behavior-changing.
+
+## 0.11.12
+  * Number formatting now accepts "int-like" inputs that don't have to
+    conform to the num interface. In particular, you can now pass an Int64
+    from the fixnum package and format it. In addition, this no longer
+    multiplies the result, so it won't lose precision on a few additional
+    cases in JS.
+
+## 0.11.11
+  * Add a -no-embedded-plurals flag to reject plurals and genders that
+    have either leading or trailing text around them. This follows the
+    ICU recommendation that a plural or gender should contain the
+    entire phrase/sentence, not just part of it.
+
+## 0.11.10
+  * Fix some style glitches with naming. The only publicly visible one
+    is DateFormat.parseUtc, but the parseUTC variant is still retained
+    for backward-compatibility.
+
+  * Provide a better error message when generating translated versions
+    and the name of a variable substitution in the message doesn't
+    match the name in the translation.
+
+## 0.11.9
+  * Fix bug with per-mille parsing (only divided by 100, not 1000)
+
+  * Support percent and per-mille formats with both positive and negative
+    variations. Previously would throw an exception for too many modifiers.
+
+## 0.11.8
+
+  * Support NumberFormats with two different grouping sizes, e.g.
+    1,23,45,67,890
+
+## 0.11.7
+  * Moved petitparser into a regular dependency so pub run works.
+
+  * Improved code layout of the package.
+
+  * Added a DateFormat.parseStrict method that rejects DateTimes with invalid
+    values and requires it to be the whole string.
+
+## 0.11.6
+
+  * Catch analyzer errors and do not generate messages for that file. Previously
+    this would stop the message extraction on syntax errors and not give error
+    messages as good as the compiler would produce. Just let the compiler do it.
+
+## 0.11.5
+
+ * Change to work with both petitparser 1.1.x and 1.2.x versions.
+
+## 0.11.4
+
+ * Broaden the pubspec constraints to allow current analyzer versions.
+
+## 0.11.3
+
+ * Add a --[no]-use-deferred-loading flag to generate_from_arb.dart and
+   generally make the deferred loading of message libraries optional.
+
+## 0.11.2
+
+ * Missed canonicalization of locales in one place in message library generation.
+
+ * Added a simple debug script for message_extraction_test.
+
+## 0.11.1
+
+ * Negative numbers were being parsed as positive.
+
+## 0.11.0
+
+ * Switch the message format from a custom JSON format to
+   the ARB format ( https://code.google.com/p/arb/ )
+
+## 0.10.0
+
+ * Make message catalogs use deferred loading.
+
+ * Update CLDR Data to version 25 for dates and numbers.
+
+ * Update analyzer dependency to allow later versions.
+
+ * Adds workaround for flakiness in DateTime creation, removes debugging code
+   associated with that.
+
+## 0.9.9
+
+* Add NumberFormat.parse()
+
+* Allow NumberFormat constructor to take an optional currency name/symbol, so
+  you can format for a particular locale without it dictating the currency, and
+  also supply the currency symbols which we don't have yet.
+
+* Canonicalize locales more consistently, avoiding a number of problems if you
+  use a non-canonical form.
+
+* For locales whose length is longer than 6 change "-" to "_" in position 3 when
+  canonicalizing. Previously anything of length > 6 was left completely alone.
+
+## 0.9.8
+
+* Add a "meaning" optional parameter for Intl.message to distinguish between
+  two messages with identical text.
+
+* Handle two different messages with the same text.
+
+* Allow complex string literals in arguments (e.g. multi-line)