* Clearly mention that netx is not maintained anymore

Modify the documentation and the runtim...

Implement padding for DoT and DoH

1. my understanding of the RFC is that the whole query must be
a multiple of the desired size, s...

According to RFC8467.

Close #95.

1. this code cannot stay in this repository because it is poised
to change quite frequently alo...

Closes #127. Related to https://github.com/ooni/probe-engine/issues/2.

Detach porcelain from netx/internal

It makes sense to discuss the scoreboard, because this is a
functionality that we hope will even...

It was not always made obvious enough.

We were previously wrapping the error in porcelain, which was not
100% correct, because technica...

This is an internal implementation detail currently used by x/porcelain. We
are planning on fork...

* ooniresolver_test.go: clarify when this test may fail

It certainly fails in my house. I hav...

It's more flexible to depend on an apex/log like interface.

cURL like logs, support Accept{,-Language}, disable auto-followup for bogons

Closes #137

Closes #113

In principle, we may want to measure a website from a place where it has
a bogon address, so we ...

porcelain: add proxy and byte counting support

Closes #133

Closes #131

Closes #130

This helps with analysing tests for SNI triggered RST injection.

* README.md: slightly change the wording

* Update README.md

* http: always enforce a body snap size

If the user explicitly requests for unbounded bodies,...

This is for clarity. Noticed when integrating with probe-engine.

It otherwise clashes with ooni/probe-engine/model and hence it's
a bit confusing when you quick...

No real API change. What changes is the way in which I think
about next, so updated the documen...

Related to #9 but implemented differently.

Closes #20

The first error is the one that is most likely to be caused by the network. Subsequent errors ar...

Complement the previous path by forcing no keepalives if the
user requests doh with a handler.
...

As explained in the code, this allows us to avoid worrying about
closing the connections that a...

Not necessarily needed but very helpful to implement OONI tests.

Xref: https://github.com/oon...

* errwrapper: map the TLS handshake timeout error

* ooniresolver: correctly recognize and han...

Xref:
- https://github.com/ooni/probe-engine/issues/87
- https://github.com/ooni/probe-engine/...

Functional to enable follow-up measurements.

Xref: https://github.com/ooni/probe-engine/issue...

Now, rather than using nervousresolver as an opt-in that is not
so easy to use, you can chain a...

While there make sure netx.go has 100% coverage

* Make sure every toplevel op is cancellable

* Significantly reduce risk of flaky tests

* Make sure all packages have 100% coverage

* porcelain: add test for late write on channel

While there, zap existing clients that were creating significant
test burden and replace them...

* http: change body reading strategy

Always read snapshots of all bodies. Limit snapshot to 1...

We were creating a system resolver that was not correctly wrapped
using the emittingresolver, h...

* internal: let the resolver emit resolve events

This fixes a bug where a standalone Resolver...

There is no actual resolve of IPs. We include the TxID also in
the connect event, so there's ac...

* Add porcelain to facilitate SNI injection testing

* Repair broken tests

Package flags mu...

We should probably introduce a longer chain of fallbacks in the
nervous resolver, but, for now,...

Not immediately necessary, but nice to have.

I've determined that it's better to add more porcelain here as
an extension, and keep ooni/prob...

This diff implements easy handling of round trips by putting into
the same event all the inform...

We need to observe real headers, so reduce the magic performed
under the hood by Go. The really...

* Rework nervous resolver and make it opt-in

I realised it was not a good idea to have it hid...

Implement retries for queries. Tested with unit testing as well
as configuring ooni/jafar to dr...

For now let us limit to act if we see any bogon. If we see at
least one, we cannot trust the up...

* Add some porcelain to simplify using netx

* fix embarrassing bug

* much better tests for...

More confidence, less stress.

The scoreboard package will probably benefit from some more
precise unit testing with respect t...

Implement a scoreboard and write there what we see. The follow up
action may or may not be requ...

I will soon use this feature to implement interim resolves info.

I am not 100% sold yet on what is the actionable top-level event
to in order to clarify in the l...

A MeasurementRoot can contain an alternative LookupHost function
which may be used to bypass th...

After trial and error, I have determined that, in general, to decide
whether we need follow-up ...

* Make sure errors are MK compatible

* we now require Go v1.13

Mainly naming suggestions by @hellais. Also, mention context
rooted measurements as an alternat...

1. make sure that, when we start a DoH transaction to resolve a
domain name, we know which di...

* Introduce pretty logging of events

It turns out I'm much more efficient at reading such log...

This is useful to distinguish the case where the error was in
sending the request from the one ...

Lots of stuff may happen between the beginning and the end of a
resolver, therefore, it makes s...

No hurry to do that externally, for now. Paves the way for
doing that quite easily, however.

I'm starting to think it would be better to pass the handler
through a context, and to have a u...

The round trip is what happens since we start thinking about
sending a request until when we re...

We are using the local port and the protocol. This makes it
possible to join HTTP and network l...

* http: start splitting the httptransport

* Drop support for Go v1.11

It does not have Clo...

* resolver: move configuration-like code to internal

* internal.go: forgot to simplify using ...

Now everything inside dialer more or less makes sense and we
have got rid of a good amount of s...

What remains code support for version 0.1.0's API

A dial may consist of more than a single connect, so it is
actually wrong to use the same ID fo...