mirror of
https://github.com/octoleo/restic.git
synced 2024-11-13 08:46:33 +00:00
452 lines
12 KiB
Markdown
452 lines
12 KiB
Markdown
|
_March 17, 2017_
|
||
|
|
||
|
Breaking Pubsub changes.
|
||
|
* Publish is now asynchronous
|
||
|
([announcement](https://groups.google.com/d/topic/google-api-go-announce/aaqRDIQ3rvU/discussion)).
|
||
|
* Subscription.Pull replaced by Subscription.Receive, which takes a callback ([announcement](https://groups.google.com/d/topic/google-api-go-announce/8pt6oetAdKc/discussion)).
|
||
|
* Message.Done replaced with Message.Ack and Message.Nack.
|
||
|
|
||
|
_February 14, 2017_
|
||
|
|
||
|
Release of a client library for Spanner. See
|
||
|
the
|
||
|
[blog post](https://cloudplatform.googleblog.com/2017/02/introducing-Cloud-Spanner-a-global-database-service-for-mission-critical-applications.html).
|
||
|
|
||
|
Note that although the Spanner service is beta, the Go client library is alpha.
|
||
|
|
||
|
_December 12, 2016_
|
||
|
|
||
|
Beta release of BigQuery, DataStore, Logging and Storage. See the
|
||
|
[blog post](https://cloudplatform.googleblog.com/2016/12/announcing-new-google-cloud-client.html).
|
||
|
|
||
|
Also, BigQuery now supports structs. Read a row directly into a struct with
|
||
|
`RowIterator.Next`, and upload a row directly from a struct with `Uploader.Put`.
|
||
|
You can also use field tags. See the [package documentation][cloud-bigquery-ref]
|
||
|
for details.
|
||
|
|
||
|
_December 5, 2016_
|
||
|
|
||
|
More changes to BigQuery:
|
||
|
|
||
|
* The `ValueList` type was removed. It is no longer necessary. Instead of
|
||
|
```go
|
||
|
var v ValueList
|
||
|
... it.Next(&v) ..
|
||
|
```
|
||
|
use
|
||
|
|
||
|
```go
|
||
|
var v []Value
|
||
|
... it.Next(&v) ...
|
||
|
```
|
||
|
|
||
|
* Previously, repeatedly calling `RowIterator.Next` on the same `[]Value` or
|
||
|
`ValueList` would append to the slice. Now each call resets the size to zero first.
|
||
|
|
||
|
* Schema inference will infer the SQL type BYTES for a struct field of
|
||
|
type []byte. Previously it inferred STRING.
|
||
|
|
||
|
* The types `uint`, `uint64` and `uintptr` are no longer supported in schema
|
||
|
inference. BigQuery's integer type is INT64, and those types may hold values
|
||
|
that are not correctly represented in a 64-bit signed integer.
|
||
|
|
||
|
* The SQL types DATE, TIME and DATETIME are now supported. They correspond to
|
||
|
the `Date`, `Time` and `DateTime` types in the new `cloud.google.com/go/civil`
|
||
|
package.
|
||
|
|
||
|
_November 17, 2016_
|
||
|
|
||
|
Change to BigQuery: values from INTEGER columns will now be returned as int64,
|
||
|
not int. This will avoid errors arising from large values on 32-bit systems.
|
||
|
|
||
|
_November 8, 2016_
|
||
|
|
||
|
New datastore feature: datastore now encodes your nested Go structs as Entity values,
|
||
|
instead of a flattened list of the embedded struct's fields.
|
||
|
This means that you may now have twice-nested slices, eg.
|
||
|
```go
|
||
|
type State struct {
|
||
|
Cities []struct{
|
||
|
Populations []int
|
||
|
}
|
||
|
}
|
||
|
```
|
||
|
|
||
|
See [the announcement](https://groups.google.com/forum/#!topic/google-api-go-announce/79jtrdeuJAg) for
|
||
|
more details.
|
||
|
|
||
|
_November 8, 2016_
|
||
|
|
||
|
Breaking changes to datastore: contexts no longer hold namespaces; instead you
|
||
|
must set a key's namespace explicitly. Also, key functions have been changed
|
||
|
and renamed.
|
||
|
|
||
|
* The WithNamespace function has been removed. To specify a namespace in a Query, use the Query.Namespace method:
|
||
|
```go
|
||
|
q := datastore.NewQuery("Kind").Namespace("ns")
|
||
|
```
|
||
|
|
||
|
* All the fields of Key are exported. That means you can construct any Key with a struct literal:
|
||
|
```go
|
||
|
k := &Key{Kind: "Kind", ID: 37, Namespace: "ns"}
|
||
|
```
|
||
|
|
||
|
* As a result of the above, the Key methods Kind, ID, d.Name, Parent, SetParent and Namespace have been removed.
|
||
|
|
||
|
* `NewIncompleteKey` has been removed, replaced by `IncompleteKey`. Replace
|
||
|
```go
|
||
|
NewIncompleteKey(ctx, kind, parent)
|
||
|
```
|
||
|
with
|
||
|
```go
|
||
|
IncompleteKey(kind, parent)
|
||
|
```
|
||
|
and if you do use namespaces, make sure you set the namespace on the returned key.
|
||
|
|
||
|
* `NewKey` has been removed, replaced by `NameKey` and `IDKey`. Replace
|
||
|
```go
|
||
|
NewKey(ctx, kind, name, 0, parent)
|
||
|
NewKey(ctx, kind, "", id, parent)
|
||
|
```
|
||
|
with
|
||
|
```go
|
||
|
NameKey(kind, name, parent)
|
||
|
IDKey(kind, id, parent)
|
||
|
```
|
||
|
and if you do use namespaces, make sure you set the namespace on the returned key.
|
||
|
|
||
|
* The `Done` variable has been removed. Replace `datastore.Done` with `iterator.Done`, from the package `google.golang.org/api/iterator`.
|
||
|
|
||
|
* The `Client.Close` method will have a return type of error. It will return the result of closing the underlying gRPC connection.
|
||
|
|
||
|
See [the announcement](https://groups.google.com/forum/#!topic/google-api-go-announce/hqXtM_4Ix-0) for
|
||
|
more details.
|
||
|
|
||
|
_October 27, 2016_
|
||
|
|
||
|
Breaking change to bigquery: `NewGCSReference` is now a function,
|
||
|
not a method on `Client`.
|
||
|
|
||
|
New bigquery feature: `Table.LoaderFrom` now accepts a `ReaderSource`, enabling
|
||
|
loading data into a table from a file or any `io.Reader`.
|
||
|
|
||
|
_October 21, 2016_
|
||
|
|
||
|
Breaking change to pubsub: removed `pubsub.Done`.
|
||
|
|
||
|
Use `iterator.Done` instead, where `iterator` is the package
|
||
|
`google.golang.org/api/iterator`.
|
||
|
|
||
|
_October 19, 2016_
|
||
|
|
||
|
Breaking changes to cloud.google.com/go/bigquery:
|
||
|
|
||
|
* Client.Table and Client.OpenTable have been removed.
|
||
|
Replace
|
||
|
```go
|
||
|
client.OpenTable("project", "dataset", "table")
|
||
|
```
|
||
|
with
|
||
|
```go
|
||
|
client.DatasetInProject("project", "dataset").Table("table")
|
||
|
```
|
||
|
|
||
|
* Client.CreateTable has been removed.
|
||
|
Replace
|
||
|
```go
|
||
|
client.CreateTable(ctx, "project", "dataset", "table")
|
||
|
```
|
||
|
with
|
||
|
```go
|
||
|
client.DatasetInProject("project", "dataset").Table("table").Create(ctx)
|
||
|
```
|
||
|
|
||
|
* Dataset.ListTables have been replaced with Dataset.Tables.
|
||
|
Replace
|
||
|
```go
|
||
|
tables, err := ds.ListTables(ctx)
|
||
|
```
|
||
|
with
|
||
|
```go
|
||
|
it := ds.Tables(ctx)
|
||
|
for {
|
||
|
table, err := it.Next()
|
||
|
if err == iterator.Done {
|
||
|
break
|
||
|
}
|
||
|
if err != nil {
|
||
|
// TODO: Handle error.
|
||
|
}
|
||
|
// TODO: use table.
|
||
|
}
|
||
|
```
|
||
|
|
||
|
* Client.Read has been replaced with Job.Read, Table.Read and Query.Read.
|
||
|
Replace
|
||
|
```go
|
||
|
it, err := client.Read(ctx, job)
|
||
|
```
|
||
|
with
|
||
|
```go
|
||
|
it, err := job.Read(ctx)
|
||
|
```
|
||
|
and similarly for reading from tables or queries.
|
||
|
|
||
|
* The iterator returned from the Read methods is now named RowIterator. Its
|
||
|
behavior is closer to the other iterators in these libraries. It no longer
|
||
|
supports the Schema method; see the next item.
|
||
|
Replace
|
||
|
```go
|
||
|
for it.Next(ctx) {
|
||
|
var vals ValueList
|
||
|
if err := it.Get(&vals); err != nil {
|
||
|
// TODO: Handle error.
|
||
|
}
|
||
|
// TODO: use vals.
|
||
|
}
|
||
|
if err := it.Err(); err != nil {
|
||
|
// TODO: Handle error.
|
||
|
}
|
||
|
```
|
||
|
with
|
||
|
```
|
||
|
for {
|
||
|
var vals ValueList
|
||
|
err := it.Next(&vals)
|
||
|
if err == iterator.Done {
|
||
|
break
|
||
|
}
|
||
|
if err != nil {
|
||
|
// TODO: Handle error.
|
||
|
}
|
||
|
// TODO: use vals.
|
||
|
}
|
||
|
```
|
||
|
Instead of the `RecordsPerRequest(n)` option, write
|
||
|
```go
|
||
|
it.PageInfo().MaxSize = n
|
||
|
```
|
||
|
Instead of the `StartIndex(i)` option, write
|
||
|
```go
|
||
|
it.StartIndex = i
|
||
|
```
|
||
|
|
||
|
* ValueLoader.Load now takes a Schema in addition to a slice of Values.
|
||
|
Replace
|
||
|
```go
|
||
|
func (vl *myValueLoader) Load(v []bigquery.Value)
|
||
|
```
|
||
|
with
|
||
|
```go
|
||
|
func (vl *myValueLoader) Load(v []bigquery.Value, s bigquery.Schema)
|
||
|
```
|
||
|
|
||
|
|
||
|
* Table.Patch is replace by Table.Update.
|
||
|
Replace
|
||
|
```go
|
||
|
p := table.Patch()
|
||
|
p.Description("new description")
|
||
|
metadata, err := p.Apply(ctx)
|
||
|
```
|
||
|
with
|
||
|
```go
|
||
|
metadata, err := table.Update(ctx, bigquery.TableMetadataToUpdate{
|
||
|
Description: "new description",
|
||
|
})
|
||
|
```
|
||
|
|
||
|
* Client.Copy is replaced by separate methods for each of its four functions.
|
||
|
All options have been replaced by struct fields.
|
||
|
|
||
|
* To load data from Google Cloud Storage into a table, use Table.LoaderFrom.
|
||
|
|
||
|
Replace
|
||
|
```go
|
||
|
client.Copy(ctx, table, gcsRef)
|
||
|
```
|
||
|
with
|
||
|
```go
|
||
|
table.LoaderFrom(gcsRef).Run(ctx)
|
||
|
```
|
||
|
Instead of passing options to Copy, set fields on the Loader:
|
||
|
```go
|
||
|
loader := table.LoaderFrom(gcsRef)
|
||
|
loader.WriteDisposition = bigquery.WriteTruncate
|
||
|
```
|
||
|
|
||
|
* To extract data from a table into Google Cloud Storage, use
|
||
|
Table.ExtractorTo. Set fields on the returned Extractor instead of
|
||
|
passing options.
|
||
|
|
||
|
Replace
|
||
|
```go
|
||
|
client.Copy(ctx, gcsRef, table)
|
||
|
```
|
||
|
with
|
||
|
```go
|
||
|
table.ExtractorTo(gcsRef).Run(ctx)
|
||
|
```
|
||
|
|
||
|
* To copy data into a table from one or more other tables, use
|
||
|
Table.CopierFrom. Set fields on the returned Copier instead of passing options.
|
||
|
|
||
|
Replace
|
||
|
```go
|
||
|
client.Copy(ctx, dstTable, srcTable)
|
||
|
```
|
||
|
with
|
||
|
```go
|
||
|
dst.Table.CopierFrom(srcTable).Run(ctx)
|
||
|
```
|
||
|
|
||
|
* To start a query job, create a Query and call its Run method. Set fields
|
||
|
on the query instead of passing options.
|
||
|
|
||
|
Replace
|
||
|
```go
|
||
|
client.Copy(ctx, table, query)
|
||
|
```
|
||
|
with
|
||
|
```go
|
||
|
query.Run(ctx)
|
||
|
```
|
||
|
|
||
|
* Table.NewUploader has been renamed to Table.Uploader. Instead of options,
|
||
|
configure an Uploader by setting its fields.
|
||
|
Replace
|
||
|
```go
|
||
|
u := table.NewUploader(bigquery.UploadIgnoreUnknownValues())
|
||
|
```
|
||
|
with
|
||
|
```go
|
||
|
u := table.NewUploader(bigquery.UploadIgnoreUnknownValues())
|
||
|
u.IgnoreUnknownValues = true
|
||
|
```
|
||
|
|
||
|
_October 10, 2016_
|
||
|
|
||
|
Breaking changes to cloud.google.com/go/storage:
|
||
|
|
||
|
* AdminClient replaced by methods on Client.
|
||
|
Replace
|
||
|
```go
|
||
|
adminClient.CreateBucket(ctx, bucketName, attrs)
|
||
|
```
|
||
|
with
|
||
|
```go
|
||
|
client.Bucket(bucketName).Create(ctx, projectID, attrs)
|
||
|
```
|
||
|
|
||
|
* BucketHandle.List replaced by BucketHandle.Objects.
|
||
|
Replace
|
||
|
```go
|
||
|
for query != nil {
|
||
|
objs, err := bucket.List(d.ctx, query)
|
||
|
if err != nil { ... }
|
||
|
query = objs.Next
|
||
|
for _, obj := range objs.Results {
|
||
|
fmt.Println(obj)
|
||
|
}
|
||
|
}
|
||
|
```
|
||
|
with
|
||
|
```go
|
||
|
iter := bucket.Objects(d.ctx, query)
|
||
|
for {
|
||
|
obj, err := iter.Next()
|
||
|
if err == iterator.Done {
|
||
|
break
|
||
|
}
|
||
|
if err != nil { ... }
|
||
|
fmt.Println(obj)
|
||
|
}
|
||
|
```
|
||
|
(The `iterator` package is at `google.golang.org/api/iterator`.)
|
||
|
|
||
|
Replace `Query.Cursor` with `ObjectIterator.PageInfo().Token`.
|
||
|
|
||
|
Replace `Query.MaxResults` with `ObjectIterator.PageInfo().MaxSize`.
|
||
|
|
||
|
|
||
|
* ObjectHandle.CopyTo replaced by ObjectHandle.CopierFrom.
|
||
|
Replace
|
||
|
```go
|
||
|
attrs, err := src.CopyTo(ctx, dst, nil)
|
||
|
```
|
||
|
with
|
||
|
```go
|
||
|
attrs, err := dst.CopierFrom(src).Run(ctx)
|
||
|
```
|
||
|
|
||
|
Replace
|
||
|
```go
|
||
|
attrs, err := src.CopyTo(ctx, dst, &storage.ObjectAttrs{ContextType: "text/html"})
|
||
|
```
|
||
|
with
|
||
|
```go
|
||
|
c := dst.CopierFrom(src)
|
||
|
c.ContextType = "text/html"
|
||
|
attrs, err := c.Run(ctx)
|
||
|
```
|
||
|
|
||
|
* ObjectHandle.ComposeFrom replaced by ObjectHandle.ComposerFrom.
|
||
|
Replace
|
||
|
```go
|
||
|
attrs, err := dst.ComposeFrom(ctx, []*storage.ObjectHandle{src1, src2}, nil)
|
||
|
```
|
||
|
with
|
||
|
```go
|
||
|
attrs, err := dst.ComposerFrom(src1, src2).Run(ctx)
|
||
|
```
|
||
|
|
||
|
* ObjectHandle.Update's ObjectAttrs argument replaced by ObjectAttrsToUpdate.
|
||
|
Replace
|
||
|
```go
|
||
|
attrs, err := obj.Update(ctx, &storage.ObjectAttrs{ContextType: "text/html"})
|
||
|
```
|
||
|
with
|
||
|
```go
|
||
|
attrs, err := obj.Update(ctx, storage.ObjectAttrsToUpdate{ContextType: "text/html"})
|
||
|
```
|
||
|
|
||
|
* ObjectHandle.WithConditions replaced by ObjectHandle.If.
|
||
|
Replace
|
||
|
```go
|
||
|
obj.WithConditions(storage.Generation(gen), storage.IfMetaGenerationMatch(mgen))
|
||
|
```
|
||
|
with
|
||
|
```go
|
||
|
obj.Generation(gen).If(storage.Conditions{MetagenerationMatch: mgen})
|
||
|
```
|
||
|
|
||
|
Replace
|
||
|
```go
|
||
|
obj.WithConditions(storage.IfGenerationMatch(0))
|
||
|
```
|
||
|
with
|
||
|
```go
|
||
|
obj.If(storage.Conditions{DoesNotExist: true})
|
||
|
```
|
||
|
|
||
|
* `storage.Done` replaced by `iterator.Done` (from package `google.golang.org/api/iterator`).
|
||
|
|
||
|
_October 6, 2016_
|
||
|
|
||
|
Package preview/logging deleted. Use logging instead.
|
||
|
|
||
|
_September 27, 2016_
|
||
|
|
||
|
Logging client replaced with preview version (see below).
|
||
|
|
||
|
_September 8, 2016_
|
||
|
|
||
|
* New clients for some of Google's Machine Learning APIs: Vision, Speech, and
|
||
|
Natural Language.
|
||
|
|
||
|
* Preview version of a new [Stackdriver Logging][cloud-logging] client in
|
||
|
[`cloud.google.com/go/preview/logging`](https://godoc.org/cloud.google.com/go/preview/logging).
|
||
|
This client uses gRPC as its transport layer, and supports log reading, sinks
|
||
|
and metrics. It will replace the current client at `cloud.google.com/go/logging` shortly.
|
||
|
|