Adding capnproto serialization to an existing Go project used to mean writing a lot of boilerplate.
Not anymore.
Given a set of golang (Go) source files, bambam will generate a capnproto schema. Even better: bambam will also generate translation functions to readily convert between your golang structs and the new capnproto structs.
You'll need a recent (up-to-date) version of go-capnproto. If you installed go-capnproto before, you'll want to update it [>= f9f239fc7f5ad9611cf4e88b10080a4b47c3951d / 16 Nov 2014].
Capnproto and go-capnproto should both be installed and on your PATH.
to install: run make. This lets us record the git commit in LASTGITCOMMITHASH to provide accurate version info. Otherwise you'll get an 'undefined: LASTGITCOMMITHASH' failure.
# be sure go-capnproto and capnpc are installed first.
$ go get -t github.com/glycerine/bambam # the -t pulls in the test dependencies.
# ignore the initial compile error about 'undefined: LASTGITCOMMITHASH'. `make` will fix that.
$ cd $GOPATH/src/github.com/glycerine/bambam
$ make # runs tests, build if all successful
$ go install
use: bambam -o outdir -p package myGoSourceFile.go myGoSourceFile2.go ...
# Bambam makes it easy to use Capnproto serialization[1] from Go.
# Bambam reads .go files and writes a .capnp schema and Go bindings.
# options:
# -o="odir" specifies the directory to write to (created if need be).
# -p="main" specifies the package header to write (e.g. main, mypkg).
# -X exports private fields of Go structs. Default only maps public fields.
# -version shows build version with git commit hash
# -OVERWRITE modify .go files in-place, adding capid tags (write to -o dir by default).
# required: at least one .go source file for struct definitions. Must be last, after options.
#
# [1] https://github.com/glycerine/go-capnproto
See rw.go.txt. To see all the files compiled together in one project: (a) comment out the defer in the rw_test.go file; (b) run go test
; (c) then cd testdir_*
and look at the sample project files there. (d). run go build
in the testdir_ to rebuild the binary. Notice that you will need all three .go files to successfully build. The two .capnp files should be kept so you can read your data from any capnp-supported language. Here's what is what in that example directory:
rw.go # your original go source file (in this test)
translateCapn.go # generated by bambam after reading rw.go
schema.capnp # generated by bambam after reading rw.go
schema.capnp.go # generated by `capnpc -ogo schema.capnp` <- you have to do this yourself or in your Makefile.
go.capnp # always necessary boilerplate to let capnpc work, just copy it from bambam/go.capnp to your build dir.
example:
jaten@c03:~/go/src/github.com/glycerine/bambam:master$ cd testdir_884497362/
jaten@c03:~/go/src/github.com/glycerine/bambam/testdir_884497362:master$ go build
jaten@c03:~/go/src/github.com/glycerine/bambam/testdir_884497362:master$ ls
go.capnp rw.go rw.go.txt schema.capnp schema.capnp.go testdir_884497362 translateCapn.go
jaten@c03:~/go/src/github.com/glycerine/bambam/testdir_884497362:master$ ./testdir_884497362
Load() data matched Saved() data.
jaten@c03:~/go/src/github.com/glycerine/bambam/testdir_884497362:master$ # run was successful
Here is what it looks like to use the Save()/Load() methods. You end up with a Save() and Load() function for each of your structs. Simple.
package main
import (
"bytes"
)
//
// By default bambam will add the `capid` tags
// to a copy of your source in the output directory.
// Use bambam -OVERWRITE to modify files directly in-place.
// The capid tags control the @0, @1, field numbering
// in the generated capnproto schema. If you change
// your go structs, the capid tags let your schema
// stay backwards compatible with prior serializations.
//
type MyStruct struct {
Hello []string `capid:"0"`
World []int `capid:"1"`
}
func main() {
rw := MyStruct{
Hello: []string{"one", "two", "three"},
World: []int{1, 2, 3},
}
// any io.ReadWriter will work here (os.File, etc)
var o bytes.Buffer
rw.Save(&o)
// now we have saved!
rw2 := &MyStruct{}
rw2.Load(&o)
// now we have restored!
}
Supported: structs, slices, and primitive/scalar types are supported. Structs that contain structs are supported. You have both slices of scalars (e.g. []int
) and slices of structs (e.g. []MyStruct
) available.
We handle [][]T
, but not [][][]T
, where T
is a struct or primitive type. The need for triply nested slices is expected to be rare. Interpose a struct after two slices if you need to go deeper.
Currently unsupported (pull requests welcome): Go maps.
Also: pointers to structs to be serialized work, but pointers in the inner-most struct do not. This is not a big limitation, as it is rarely meaningful to pass a pointer value to a different process.
When you run bambam
, it will generate a modified copy of your go source files in the output directory.
These new versions include capid tags on all public fields of structs. You should inspect the copy of the source file in the output directory, and then replace your original source with the tagged version. You can also manually add capid tags to fields, if you need to manually specify a field number (e.g. you are matching an pre-existing capnproto definition).
If you are feeling especially bold, bambam -OVERWRITE my.go
will replace my.go with the capid tagged version. For safety, only do this on backed-up and version controlled source files.
By default only public fields (with a Capital first letter in their name) are tagged. The -X flag ignores the public/private distinction, and tags all fields.
The capid tags allow the capnproto schema evolution to function properly as you add new fields to structs. If you don't include the capid tags, your serialization code won't be backwards compatible as you change your structs.
Deleting fields from your go structs isn't (currently) particularly well-supported. We could potentially allow fields to be // commented out in the go source and yet still parse the comments and use that parse to keep the schema correct, but that's not a trivial bit of work.
example of capid annotion use
type Job struct {
C int `capid:"2"` // we added C later, thus it is numbered higher.
A int `capid:"0"`
B int `capid:"1"`
}
Also available tags: capid:"skip"
or capid:"-1"
(any negative number): this field will be skipped and not serialized or written to the schema.
// capname:"Counter"
type number struct {
A int
}
The above struct will be mapped into capnproto as:
struct Counter {
a @0: Int64;
}
Without the // capname:"Counter"
comment, you would get:
struct NumberCapn {
a @0: Int64;
}
Explanation: Using a // capname:"newName"
comment on the line right before a struct definition will cause bambam
to use 'newName' as the name for the corresponding struct in the capnproto schema. Otherwise the corresponding struct will simply uppercase the first letter of the orignal Go struct, and append "Capn". For example: a Go struct called number
would induce a parallel generated capnp struct called NumberCapn
.
see build.cmd
. Thanks to Klaus Post (http://klauspost.com) for contributing this.
Copyright (c) 2015, Jason E. Aten, Ph.D.