Files
mgmt/lang/funcs/funcs.go
James Shubin a07dc0a511 lang: interpolate, funcs: Add a concat function for faster interpolation
This adds a concat function which can be used directly by string
interpolation to avoid having to constantly unify the plus operator
which is much slower at this time.

The new monomorphisms changes caused type unification of a notable
example to go from ~25s to ~5m30s which was obviously not bearable. With
this fix, things are now down to ~6s.

This is an important optimization, but it's also a good reminder that
type unification of polymorphic functions needs to be improved in
general too.
2024-01-02 15:23:40 -05:00

319 lines
10 KiB
Go

// Mgmt
// Copyright (C) 2013-2023+ James Shubin and the project contributors
// Written by James Shubin <james@shubin.ca> and the project contributors
//
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU General Public License
// along with this program. If not, see <http://www.gnu.org/licenses/>.
// Package funcs provides a framework for functions that change over time.
package funcs
import (
"context"
"fmt"
"strings"
"sync"
"github.com/purpleidea/mgmt/lang/interfaces"
"github.com/purpleidea/mgmt/lang/types"
"github.com/purpleidea/mgmt/util/errwrap"
)
const (
// ModuleSep is the character used for the module scope separation. For
// example when using `fmt.printf` or `math.sin` this is the char used.
// It is included here for convenience when importing this package.
ModuleSep = interfaces.ModuleSep
// ReplaceChar is a special char that is used to replace ModuleSep when
// it can't be used for some reason. This currently only happens in the
// golang template library. Even with this limitation in that library,
// we don't want to allow this as the first or last character in a name.
// NOTE: the template library will panic if it is one of: .-#
ReplaceChar = "_"
// CoreDir is the directory prefix where core mcl code is embedded.
CoreDir = "core/"
// ConcatFuncName is the name the concat function is registered as. It
// is listed here because it needs a well-known name that can be used by
// the string interpolation code.
ConcatFuncName = "concat"
)
// registeredFuncs is a global map of all possible funcs which can be used. You
// should never touch this map directly. Use methods like Register instead. It
// includes implementations which also satisfy PolyFunc as well.
var registeredFuncs = make(map[string]func() interfaces.Func) // must initialize
// Register takes a func and its name and makes it available for use. It is
// commonly called in the init() method of the func at program startup. There is
// no matching Unregister function. You may also register functions which
// satisfy the PolyFunc interface. To register a function which lives in a
// module, you must join the module name to the function name with the ModuleSep
// character. It is defined as a const and is probably the period character.
func Register(name string, fn func() interfaces.Func) {
if _, exists := registeredFuncs[name]; exists {
panic(fmt.Sprintf("a func named %s is already registered", name))
}
// can't contain more than one period in a row
if strings.Index(name, ModuleSep+ModuleSep) >= 0 {
panic(fmt.Sprintf("a func named %s is invalid", name))
}
// can't start or end with a period
if strings.HasPrefix(name, ModuleSep) || strings.HasSuffix(name, ModuleSep) {
panic(fmt.Sprintf("a func named %s is invalid", name))
}
// TODO: this should be added but conflicts with our internal functions
// can't start or end with an underscore
//if strings.HasPrefix(name, ReplaceChar) || strings.HasSuffix(name, ReplaceChar) {
// panic(fmt.Sprintf("a func named %s is invalid", name))
//}
fnx := fn() // check that all functions have migrated to the new API!
if _, ok := fnx.(interfaces.OldPolyFunc); ok {
if _, ok := fnx.(interfaces.PolyFunc); !ok {
panic(fmt.Sprintf("a func named %s implements OldPolyFunc but not PolyFunc", name))
}
}
//gob.Register(fn())
registeredFuncs[name] = fn
}
// ModuleRegister is exactly like Register, except that it registers within a
// named module.
func ModuleRegister(module, name string, fn func() interfaces.Func) {
Register(module+ModuleSep+name, fn)
}
// Lookup returns a pointer to the function's struct. It may be convertible to a
// PolyFunc if the particular function implements those additional methods.
func Lookup(name string) (interfaces.Func, error) {
f, exists := registeredFuncs[name]
if !exists {
return nil, fmt.Errorf("not found")
}
return f(), nil
}
// LookupPrefix returns a map of names to functions that start with a module
// prefix. This search automatically adds the period separator. So if you want
// functions in the `fmt` package, search for `fmt`, not `fmt.` and it will find
// all the correctly registered functions. This removes that prefix from the
// result in the map keys that it returns. If you search for an empty prefix,
// then this will return all the top-level functions that aren't in a module.
func LookupPrefix(prefix string) map[string]func() interfaces.Func {
result := make(map[string]func() interfaces.Func)
for name, f := range registeredFuncs {
// requested top-level functions, and no module separators...
if prefix == "" {
if !strings.Contains(name, ModuleSep) {
result[name] = f // copy
}
continue
}
sep := prefix + ModuleSep
if !strings.HasPrefix(name, sep) {
continue
}
s := strings.TrimPrefix(name, sep) // remove the prefix
result[s] = f // copy
}
return result
}
// Map returns a map from all registered function names to a function to return
// that one. We return a copy of our internal registered function store so that
// this result can be manipulated safely. We return the functions that produce
// the Func interface because we might use this result to create multiple
// functions, and each one must have its own unique memory address to work
// properly.
func Map() map[string]func() interfaces.Func {
m := make(map[string]func() interfaces.Func)
for name, fn := range registeredFuncs { // copy
m[name] = fn
}
return m
}
// PureFuncExec is usually used to provisionally speculate about the result of a
// pure function, by running it once, and returning the result. Pure functions
// are expected to only produce one value that depends only on the input values.
// This won't run any slow functions either.
func PureFuncExec(handle interfaces.Func, args []types.Value) (types.Value, error) {
hostname := "" // XXX: add to interface
debug := false // XXX: add to interface
logf := func(format string, v ...interface{}) {} // XXX: add to interface
ctx, cancel := context.WithCancel(context.TODO())
defer cancel()
info := handle.Info()
if !info.Pure {
return nil, fmt.Errorf("func is not pure")
}
// if function is expensive to run, we won't run it provisionally
if info.Slow {
return nil, fmt.Errorf("func is slow")
}
if err := handle.Validate(); err != nil {
return nil, errwrap.Wrapf(err, "could not validate func")
}
sig := handle.Info().Sig
if sig.Kind != types.KindFunc {
return nil, fmt.Errorf("must be kind func")
}
ord := handle.Info().Sig.Ord
if i, j := len(ord), len(args); i != j {
return nil, fmt.Errorf("expected %d args, got %d", i, j)
}
wg := &sync.WaitGroup{}
defer wg.Wait()
errch := make(chan error)
input := make(chan types.Value) // we close this when we're done
output := make(chan types.Value) // we create it, func closes it
init := &interfaces.Init{
Hostname: hostname,
Input: input,
Output: output,
World: nil, // should not be used for pure functions
Debug: debug,
Logf: func(format string, v ...interface{}) {
logf("func: "+format, v...)
},
}
if err := handle.Init(init); err != nil {
return nil, errwrap.Wrapf(err, "could not init func")
}
close1 := make(chan struct{})
close2 := make(chan struct{})
wg.Add(1)
go func() {
defer wg.Done()
defer close(errch) // last one turns out the lights
select {
case <-close1:
}
select {
case <-close2:
}
}()
wg.Add(1)
go func() {
defer wg.Done()
defer close(close1)
if debug {
logf("Running func")
}
err := handle.Stream(ctx) // sends to output chan
if debug {
logf("Exiting func")
}
if err == nil {
return
}
// we closed with an error...
select {
case errch <- errwrap.Wrapf(err, "problem streaming func"):
}
}()
wg.Add(1)
go func() {
defer wg.Done()
defer close(close2)
defer close(input) // we only send one value
if len(args) == 0 {
return
}
si := &types.Type{
// input to functions are structs
Kind: types.KindStruct,
Map: handle.Info().Sig.Map,
Ord: handle.Info().Sig.Ord,
}
st := types.NewStruct(si)
for i, arg := range args {
name := handle.Info().Sig.Ord[i]
if err := st.Set(name, arg); err != nil { // populate struct
select {
case errch <- errwrap.Wrapf(err, "struct set failure"):
}
return
}
}
select {
case input <- st: // send to function (must not block)
case <-close1: // unblock the input send in case stream closed
select {
case errch <- fmt.Errorf("stream closed early"):
}
}
}()
once := false
var result types.Value
var reterr error
Loop:
for {
select {
case value, ok := <-output: // read from channel
if !ok {
output = nil
continue Loop // only exit via errch closing!
}
if once {
reterr = fmt.Errorf("got more than one value")
continue // only exit via errch closing!
}
once = true
result = value // save value
case err, ok := <-errch: // handle possible errors
if !ok {
break Loop
}
if err == nil {
// programming error
err = fmt.Errorf("error was missing")
}
e := errwrap.Wrapf(err, "problem streaming func")
reterr = errwrap.Append(reterr, e)
}
}
cancel()
if result == nil && reterr == nil {
// programming error
// XXX: i think this can happen when we exit without error, but
// before we send one output message... not sure how this happens
// XXX: iow, we never send on output, and errch closes...
// XXX: this could happen if we send zero input args, and Stream exits without error
return nil, fmt.Errorf("function exited with nil result and nil error")
}
return result, reterr
}