- `go` spawns lightweight thread.
- `chan` and `select` provide channels with Go semantic.
- `method` allows to define methods separate from class.
- `gimport` allows to import python modules by full path in a Go workspace.

The focus of first draft was on usage interface and on correctness, not speed.
In particular select should be fully working.

If there is a chance I will maybe try to followup with gevent-based
implementation in the future.

Lonet is the virtnet network that, contrary to pipenet, could be used
when there are several OS-level processes involved.

It uses SQLite for its registry and native OS-level TCP over loopback
for data exchange. There is small text-based connection handshake
protocol prelude that have to be carried out when a connection is tried
to be established to host via its subnetwork, but after it data exchange
goes directly through OS TCP stack.

Lonet documentation follows:

"""
Package lonet provides TCP network simulated on top of localhost TCP loopback.

For testing distributed systems it is sometimes handy to imitate network of
several TCP hosts. It is also handy that ports allocated on Dial/Listen/Accept on
that hosts be predictable - that would help tests to verify network events
against expected sequence. When whole system could be imitated in 1 OS-level
process, package lab.nexedi.com/kirr/go123/xnet/pipenet serves the task via
providing TCP-like synchronous in-memory network of net.Pipes. However
pipenet cannot be used for cases where tested system consists of 2 or more
OS-level processes. This is where lonet comes into play:

Similarly to pipenet addresses on lonet are host:port pairs and several
hosts could be created with different names. A host is xnet.Networker and
so can be worked with similarly to regular TCP network access-point with
Dial/Listen/Accept. Host's ports allocation is predictable: ports of a host
are contiguous integer sequence starting from 1 that are all initially free,
and whenever autobind is requested the first free port of the host will be
used.

Internally lonet network maintains registry of hosts so that lonet
addresses could be resolved to OS-level addresses, for example α:1 and β:1
to 127.0.0.1:4567 and 127.0.0.1:8765, and once lonet connection is
established it becomes served by OS-level TCP connection over loopback.

Example:

net, err := lonet.Join(ctx, "mynet")
hα, err := net.NewHost(ctx, "α")
hβ, err := net.NewHost(ctx, "β")

// starts listening on address "α:10"
l, err := hα.Listen(":10")
go func() {
 csrv, err := l.Accept()   // csrv will have LocalAddr "α:1"
}()
ccli, err := hβ.Dial(ctx, "α:10") // ccli will be connection between "β:1" - "α:1"

Once again lonet is similar to pipenet, but since it works via OS TCP stack
it could be handy for testing networked application when there are several
OS-level processes involved.
"""

"""
Lonet organization

For every lonet network there is a registry with information about hosts
available on the network, and for each host its OS-level listening address.
The registry is kept as SQLite database under

/<tmp>/lonet/<network>/registry.db

Whenever host α needs to establish connection to address on host β, it
queries the registry for β and further talks to β on that address.
Correspondingly when a host joins the network, it announces itself to the
registry so that other hosts could see it.

Handshake protocol

After α establishes OS-level connection to β via main β address, it sends
request to further establish lonet connection on top of that:

> lonet "<network>" dial "<α:portα>" "<β:portβ>"\n

β checks whether portβ is listening, and if yes, accepts the connection on
corresponding on-β listener with giving feedback to α that connection was
accepted:

< lonet "<network>" connected "<β:portβ'>"\n

After that connection is considered to be lonet-established and all further
exchange on it is directly controlled by corresponding lonet-level
Read/Write on α and β.

If, on the other hand, lonet-level connection cannot be established, β replies:

< lonet "<networkβ>" E "<error>"\n

where <error> could be:

- connection refused if <β:portβ> is not listening
- network mismatch if β thinks it works on different lonet network than α
- protocol error if β thinks that α send incorrect dial request
- ...
"""

As we are going to implement another virtual network it would be good to
share common code between implementations. For this generalize pipenet
implementation to also cover the case when one does not own full network
and owns only some hosts of it.

An example of such situation is when one process handles one group of
virtual hosts and another process handles another group of virtual
hosts. Below a group of virtual hosts handled as part of network is
called subnetwork.

If hosts are not created in one place, we need a way to communicate
information about new hosts in between subnetworks. This leads to using
some kind of "registry" (see Registry interface).

Then for the common code to be reused by a virtual network
implementation it has to provide its working in the form of Engine
interface to that common code. In turn the common code exposes another
- Notifier - interface for particular network implementation to notify
common code of events that come from outside to the subnetwork.

Pipenet is reworked to be just a client of the common virtnet
infrastructure.

Virtnet documentation follows:

"""
Package virtnet provides infrastructure for TCP-like virtual networks.

For testing distributed systems it is sometimes handy to imitate network of
several TCP hosts. It is also handy that ports allocated on Dial/Listen/Accept
on that hosts be predictable - that would help tests to verify network
events against expected sequence.

Package virtnet provides infrastructure for using and implementing such
TCP-like virtual networks.

Using virtnet networks

Addresses on a virtnet network are host:port pairs represented by Addr.
A network conceptually consists of several SubNetworks each being home for
multiple Hosts. Host is xnet.Networker and so can be worked with similarly
to regular TCP network access-point with Dial/Listen/Accept. Host's ports
allocation is predictable: ports of a host are contiguous integer sequence
starting from 1 that are all initially free, and whenever autobind is
requested the first free port of the host will be used.
Virtnet ensures that host names are unique throughout whole network.

To work with a virtnet network, one uses corresponding package for
particular virtnet network implementation. Such packages provide a way to
join particular network and after joining give back SubNetwork to user.
Starting from SubNetwork one can create Hosts and from those exchange data
throughout whole network.

Please see package lab.nexedi.com/kirr/go123/xnet/pipenet for particular
well-known virtnet-based network.

Implementing virtnet networks

To implement a virtnet-based network one need to implement Engine and Registry.

A virtnet network implementation should provide Engine and Registry
instances to SubNetwork when creating it. The subnetwork will use provided
engine and registry for its operations. A virtnet network implementation
receives instance of Notifier together with SubNetwork when creating it. The
network implementation should use provided Notifier to notify the subnetwork
to handle incoming events.

Please see Engine, Registry and Notifier documentation for details.
"""

Another virtnet-based network that is not limited to be used only in 1
OS process will follow next.

- clarify docstrings;
- unexport NetPrefix.

On TCP port(accepted) = port(listen), i.e.  When we connect
www.nexedi.com:80, remote addr of socket will have port 80.  Likewise on
server side accepted socket will have local port 80.  The connection
should be thus fully identified by src-dst address pair.

However currently pipenet, when accepting connection, allocates another
free port at acceptor side.

Quoting documentation:

"""
Package xcontext provides addons to std package context.

Merging contexts

Merge could be handy in situations where spawned job needs to be canceled
whenever any of 2 contexts becomes done. This frequently arises with service
methods that accept context as argument, and the service itself, on another
control line, could be instructed to become non-operational. For example:

	func (srv *Service) DoSomething(ctx context.Context) (err error) {
		defer xerr.Contextf(&err, "%s: do something", srv)

       		// srv.serveCtx is context that becomes canceled when srv is
       		// instructed to stop providing service.
       		origCtx := ctx
       		ctx, cancel := xcontext.Merge(ctx, srv.serveCtx)
       		defer cancel()

       		err = srv.doJob(ctx)
       		if err != nil {
       			if ctx.Err() != nil && origCtx.Err() == nil {
       				// error due to service shutdown
       				err = ErrServiceDown
       			}
       			return err
       		}

		...
	}
"""

- clarify Networker.Name on example of NetPlain;
- we don't need Networker.Addr since we have Networker.Name;
- format Dial/Listen docstrings so that they render properly under godoc;
- format list in NetTLS doc so that it renders properly under godoc.

profile/trace outputs + files left after `go test -c`.

As of go 47be3d49 there are 2 new types in g compared to go1.10:
waitReason and ancestorInfo.

Preliminary because Go1.11 has not been released yet, so
zruntime_g_go1.11.go will need to be potentially regenerated after
the release.

With previous sed expression it was failing e.g. on runtime.guintptr,
giving much more after type definition:

	$ go doc -c -u runtime.muintptr | sed -n -e '/^type /,/^}/p'
	type muintptr uintptr
	    muintptr is a *m that is not tracked by the garbage collector.

	    Because we do free Ms, there are some additional constrains on muintptrs:

	    1. Never hold an muintptr locally across a safe point.

	    2. Any muintptr in the heap must be owned by the M itself so it can

	    ensure it is not in use when the last true *m is released.

	func (mp muintptr) ptr() *m
	func (mp *muintptr) set(m *m)

Fix it since we'll need to extract more single-line types for Go1.11 support.

To verify it works de-stub {g,p,m}uintptr.

Go1.10 now modifies CGo sources as text, not AST, and this way comments
are not removed and gotrace can see its '//trace:event ' comments in a
CGo *.go files: https://github.com/golang/go/commit/85c3ebf4.

This way with Go1.10 tests were failing because `//trace:event
traceHello()` in a/pkg1/pkg1c.go was now noticed by gotrace and more
then expected trace events produces.

I have not noticed this in 65c399f0 (tracing/runtime: Add support for
Go1.10 (preliminary)) probably because at that time above Go commit was
not in my local Go1.10 tree.

For the reference: tracing/runtime support regenerated for Go1.10.2
stays exactly the same as in 65c399f0 and so does not need updating.

Commit 79e328c5 (xerr += First, Merge) was not right saying that

	if err == io.EOF

will panic if err has dynamic type Errorv. It will not because
interfaces with different dynamic types are always not equal.
However Errorv == Errorv will indeed panic.

Document and test that it is safe to compare error vectors to other
errors, e.g. to io.EOF, and that Errorv == Errorv panics.

I was too used to my old Python habit while originally writing this.
Using the occasion add package-level description which outlines the
package to a user.

We already have TraceConnect which corresponds to event of established
network connection.

However in many test scenarios it is required to know and pause
execution right before dial is going to happen. For this introduce
TraceDial which represents event corresponding to network dialing start.

Networker.Name should return name of access-point Networker represents
on the network.

We will need this information in the next patch for tracing dial events,
when there is no connection established yet, and thus dialer location
has to be taken from somewhere. It is also generally a good idea for
Networker to have a name.

For NetPlain networker the name is local hostname.

Should be in d3a7a196 (xnet/pipenet: Package for TCP-like synchronous
in-memory network of net.Pipes)

I was too used to my old Python habit while originally writing this.
Using the occasion add package-level description which outlines the
package to a user.

Until https://github.com/golang/go/issues/7873 is fixed/implemented,
let's follow the currently advised scheme to indent lists and this way
put them as pre-formatted text.

We already do so mostly everywhere, but 2 of the places were not
following it and thus all items there were rendered as one line.

For example in ZODB FileStorage format reader routines are written
working with io.ReaderAt for the following reasons:

- for loads random-access is required,
- there can be several concurrent loads in flight simultaneously.

At the same time various database iterations (APIs additional to load)
use sequential access pattern and can be served by the same record
reading routines. However with them we cannot use e.g. bufio.Reader
because it works with plain io.Reader, not io.ReaderAt.

Here comes SeqReaderAt: it adds a buffer, by default 2·4K, on top of
original io.Reader, automatically detects direction of sequential
access which can be forward, backward, or interleaved forward-backward
patterns, and buffers data accordingly to avoid many syscalls e.g. in
os.File case.

Currently by direct & buffered readers that also can report current
offset in input stream.

This functionality is handy to have when for example one needs to report
an error after finding decoding problem and wants to include particular
stream position in the report.

Freelist of buffers is frequently needed to avoid GC/zeroing overhead in
e.g. networked servers. Making buffer reference-counted is then required
if buffer is shared between several users so that it is clear when it
can go back to its pool. In ZODB this situation was arising on returning
the same buffer from cached loads if several loads for same data are
issued in close to parallel.

[1] https://lab.nexedi.com/kirr/neo/tree/75a71514/go/zodb

Preliminary because Go1.10 has not been released yet, so
zruntime_g_go1.10.go will need to be potentially regenerated after
the release.

./gotrace.go:707: Errorf format %s arg pkgi.Pkg.Path is a func value, not called

xruntime/xruntime_test.go:47: Fatal call has possible formatting directive %v

./alloc_test.go:62: Fatalf format %d has arg aliases(s, sprev) of wrong type bool

This tool analyzes `go tool trace -d` output and take notices on
ProcStart and GoStart events to track G->P->M relation to check how
often a G changes M.

Unfortunately it seems to be a frequent event and changing M probably means
changing CPU and thus loosing CPU caches.

This patch adds pipenet - virtual network of net.Pipes.

Addresses on pipenet are host:port pairs. A host is xnet.Networker and so
can be worked with similarly to regular TCP network with Dial/Listen/Accept/...

Example:

    net := pipenet.New("")
    h1 := net.Host("abc")
    h2 := net.Host("def")

    l, err := h1.Listen(":10")       // starts listening on address "abc:10"
    go func() {
            csrv, err := l.Accept()  // csrv will have LocalAddr "abc:10"
    }()
    ccli, err := h2.Dial("abc:10")   // ccli will have RemoteAddr "def:10"

Pipenet might be handy for testing interaction of networked applications in 1
process without going to OS networking stack.

This patch adds xnet.NetTrace which wraps a networker calling
notification functions on Connect/Listen/Tx events.

The code is draft and I'm not sure adding this functionality is good
idea, but I still add it for completeness and because there is one user
for it.

Please do not expect the interface of xnet tracing to be stable.

Std net.Conn already can represent both plain TCP connections and
connections wrapped with TLS. However the connections itself need to be
created differently. This might become inconvenient when establishing
connections should be inside server logic where it is desirable to have
one codepath which works uniformly via interfaces.

This patch introduces Networker - new interface which represents
access-point to a streaming network. One can Dial or Listen on it and
get underlying network name.

Two functions are also provided to create networkers for plain TCP
and to wrap a networker with TLS layer.

This way one can initially decide and setup a networker, pass it to
server logic, and server inside uses just Networker interface
transparently either listening/connecting via regular sockets, or via
sockets wrapped with TLS layer.

XRun (added in db941f12) and Runx (added in 486ede30) run a function and
convert error <-> exception back and forth. However sometimes it is
handy to only convert a function but not run it - e.g. for passing into

	x/sync/errgroup.Group.Go

To do so this patch adds XFunc and Funcx - functional counterparts to
XRun and Runx.

No new tests are needed because now XRun and Runx are just tiny wrappers
around new functions and we already have tests for XRun and Runx.

Race-detector does not know Probe.Detach works under world stopped and
that this way it cannot break consistency of probes list attached to a
trace event - on event signalling either a probe will be run or not run
at all.

And we do not mind that e.g. while Detach was in progress a probe was
read from traceevent list and decided to be run and the probe
function was actually called just after Detach finished.

For this reason tell race-detector to not take into account all memory
read/write that are performed while the world is stopped.

If we do not it complains e.g. this way:

    ==================
    WARNING: DATA RACE
    Read at 0x00c42000d760 by goroutine 7:
      lab.nexedi.com/kirr/neo/go/zodb/storage._traceCacheGCFinish_run()
          /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/xcommon/tracing/tracing.go:265 +0x81
      lab.nexedi.com/kirr/neo/go/zodb/storage.traceCacheGCFinish()
          /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/storage/ztrace.go:22 +0x63
      lab.nexedi.com/kirr/neo/go/zodb/storage.(*Cache).gc()
          /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/storage/cache.go:497 +0x62c
      lab.nexedi.com/kirr/neo/go/zodb/storage.(*Cache).gcmain()
          /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/storage/cache.go:478 +0x4c

    Previous write at 0x00c42000d760 by goroutine 6:
      lab.nexedi.com/kirr/neo/go/xcommon/tracing.(*Probe).Detach()
          /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/xcommon/tracing/tracing.go:319 +0x103
      lab.nexedi.com/kirr/neo/go/xcommon/tracing.(*ProbeGroup).Done()
          /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/xcommon/tracing/tracing.go:344 +0xa5
      lab.nexedi.com/kirr/neo/go/zodb/storage.TestCache()
          /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/storage/cache_test.go:576 +0x7f94
      testing.tRunner()
          /home/kirr/src/tools/go/go/src/testing/testing.go:746 +0x16c

    Goroutine 7 (running) created at:
      lab.nexedi.com/kirr/neo/go/zodb/storage.NewCache()
          /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/storage/cache.go:129 +0x227
      lab.nexedi.com/kirr/neo/go/zodb/storage.TestCache()
          /home/kirr/src/neo/src/lab.nexedi.com/kirr/neo/go/zodb/storage/cache_test.go:165 +0x7b1
      testing.tRunner()
          /home/kirr/src/tools/go/go/src/testing/testing.go:746 +0x16c

    Goroutine 6 (finished) created at:
      testing.(*T).Run()
          /home/kirr/src/tools/go/go/src/testing/testing.go:789 +0x568
      testing.runTests.func1()
          /home/kirr/src/tools/go/go/src/testing/testing.go:1004 +0xa7
      testing.tRunner()
          /home/kirr/src/tools/go/go/src/testing/testing.go:746 +0x16c
      testing.runTests()
          /home/kirr/src/tools/go/go/src/testing/testing.go:1002 +0x521
      testing.(*M).Run()
          /home/kirr/src/tools/go/go/src/testing/testing.go:921 +0x206
      main.main()
          lab.nexedi.com/kirr/neo/go/zodb/storage/_test/_testmain.go:44 +0x1d3
    ==================

As it was said in the previous patch here goes gotrace utility to
process `//trace:event ...` and other tracing related directives.

Related excerpt from the documentation:

---- 8< ----
Gotrace

The way //trace:event and //trace:import works is via additional code being
generated for them. Whenever a package uses any //trace: directive,
it has to organize to run `gotrace gen` on its sources for them to work,
usually with the help of //go:generate. For example:

	package hello

	//go:generate gotrace gen .

	//trace:event ...

Besides `gotrace gen` gotrace has other subcommands also related to tracing,
for example `gotrace list` lists trace events a package provides.
---- 8< ----

Gotrace works by parsing and typechecking go sources via go/parse &
go/types and then for special comments generating corresponding
additional code that is supported by tracing runtime.

Package tracing will provide usage and runtime support for Linux-style
traceevents/tracepoints for Go.

This patch comes with the runtime support to attach/detach probes to/from
trace-events and stop/restart the world so that attaching/detaching can
be done safely while nothing else is running. Having attach/detach under
STW allows regular probe list iteration to be done without locks.

The next patch will add gotrace utility which automatically turns
//trace:event in-source comments into proper trace-event definitions.

Below is excerpt from tracing usage. Please refer to tracing.go for full
text.

---- 8< ----
Package tracing provides usage and runtime support for Go tracing facilities.

Trace events

A Go package can define several events of interest to trace via special
comments. With such definition a tracing event becomes associated with trace
function that is used to signal when the event happens. For example:

	package hello

	//trace:event traceHelloPre(who string)
	//trace:event traceHello(who string)

	func SayHello(who string) {
		traceHelloPre(who)
		fmt.Println("Hello, %s", who)
		traceHello(who)
	}

By default trace function does nothing and has very small overhead.

Probes

However it is possible to attach probing functions to events. A probe, once
attached, is called whenever event is signalled in the context which triggered
the event and pauses original code execution until the probe is finished. It is
possible to attach several probing functions to the same event and dynamically
detach/(re-)attach them at runtime. Attaching/detaching probes must be done
under tracing.Lock. For example:

	type saidHelloT struct {
		who  string
		when time.Time
	}
	saidHello := make(chan saidHelloT)

	tracing.Lock()
	p := traceHello_Attach(nil, func(who string) {
		saidHello <- saidHelloT{who, time.Now()}
	})
	tracing.Unlock()

	go func() {
		for hello := range saidHello {
			fmt.Printf("Said hello to %v @ %v\n", hello.who, hello.when)
		}
	}()

	SayHello("JP")
	SayHello("Kirr")
	SayHello("Varya")

	tracing.Lock()
	p.Detach()
	tracing.Unlock()

	close(saidHello)

For convenience it is possible to keep group of attached probes and detach them
all at once using ProbeGroup:

	pg := &tracing.ProbeGroup{}

	tracing.Lock()
	traceHelloPre_Attach(pg, func(who string) { ... })
	traceHello_Attach(pg, func(who string) { ... })
	tracing.Unlock()

	// some activity

	// when probes needs to be detached (no explicit tracing.Lock needed):
	pg.Done()

Probes is general mechanism which allows various kinds of trace events usage.
Three ways particularly are well-understood and handy:

	- recording events stream
	- profiling
	- synchronous tracing

...

Initial draft. The implementation is modelled after `git` and `go`.

4d9a613c (Relicense to GPLv3+ with wide exception for all Free Software
/ Open Source projects + Business options.) added more lines to
my/my_test.go than removed (@@ -5,16 +5,18 @@) so the line number of

        myline := Line()

changed by 2 (= 18 - 16).

Fix the test accordingly.

Relicense to GPLv3+ with wide exception for all Free Software / Open Source projects + Business options.

Nexedi stack is licensed under Free Software licenses with various exceptions
that cover three business cases:

- Free Software
- Proprietary Software
- Rebranding

As long as one intends to develop Free Software based on Nexedi stack, no
license cost is involved. Developing proprietary software based on Nexedi stack
may require a proprietary exception license. Rebranding Nexedi stack is
prohibited unless rebranding license is acquired.

Through this licensing approach, Nexedi expects to encourage Free Software
development without restrictions and at the same time create a framework for
proprietary software to contribute to the long term sustainability of the
Nexedi stack.

Please see https://www.nexedi.com/licensing for details, rationale and options.

CeilLog2(x)	= min i: 2^i >= x
FloorLog2(x)	= max i: 2^i <= x