Part of https://github.com/ooni/probe/issues/2669

We need this feature to write test for StreamAllContext, which in turn
is one of the TODOs refe...

Part of https://github.com/ooni/probe/issues/2556

The previous martian code used http/1.1 by default. Now I know why: if
using http2 we cannot hi...

It turns out we can further simplify the ooni/probe-cli implementation
by adding more methods o...

The UNetStack and CA structures have similar functionality exposed
through a different interfac...

The refactoring at
https://github.com/ooni/netem/commit/45a7aa975636a8de271e07904ccd076a953618b...

Before writing this diff, netem automatically generated the correct cert
on the fly based on th...

It should not happen because GVisor should route the traffic to itself
but it's better to be su...

Required to write some tests for
https://github.com/ooni/probe/issues/2531

Also part of https://github.com/ooni/probe/issues/1803

To move forward with https://github.com/ooni/probe/issues/1803, I want
to generate expired cert...

This diff implements the missing DPI rules required to complete
https://github.com/ooni/probe/i...

Required to move forward with https://github.com/ooni/probe/issues/1803

Part of https://github.com/ooni/probe/issues/1803

Useful to move forward with https://github.com/ooni/probe/issues/1803

Part of https://github.com/ooni/probe/issues/2029

I noticed this was possible while working on
https://github.com/ooni/probe-cli/pull/1205.

Re...

These two functionalities are currently hidden, but users of this
library need to write less du...

This diff adds support for sending a RST segment when the client to
server flow includes a spec...

Reference issue: [#2464](https://github.com/ooni/probe/issues/2464)

This diff aims to fix the...

This diff upgrades to go1.20 and a recenty development snapshot posted
on the `go` branch of gv...

This diff implements DNS spoofing. To this end, we restructure how RST
injection works first. W...

Otherwise, if the user does not provide a canonical CNAME, the server
code will panic at runtime.

So, we can model a link with more losses and more delay for the
throttled flow, which further s...

In this note, we explain why we think it's fine to use one of the link
models we have to write ...

Let's document how this Go module is structured.

This PR adds additional document that is nice to have.

* dnsping is already an integration test

* httping is also already an integration test

* c...

While there, get rid of the throttle command, which was useful earlier
on during development, b...

I think it's better to split them. Eventually coverage will be based on
unit tests while race d...

I do not fully understand what coveralls is showing and I'd like to see
the toplevel package an...

This diff attempts to make integration tests less flaky.

While there, recognize that we should support an UnderlyingNetwork
implemented using the stdlib...

While there, notice that we have a flaky test.

We were using the wrong link layer constant.

Now that we have a better link forwarding code implementation, drop the
old one.

Rather than always using the same function to emulate a link, choose the
best function by inspe...

This algorithm is an evolution of the current link forwarding algorithm
attempting to make TCP ...

We put mandatory arguments first and optional ones later. In the current
state of the main bran...

We will select this algorithm when the user only configures a delay
without any PLR or DPI need...

This diff adds support for a fast forwarding algorithm that has as
little as possible overhead....

We need to include in each row:

1. the prefix of the PCAP files we're capturing;

2. the im...

By having a dedicated flag, we can separate dropping a single packet
from imposing a specific e...

Lacking precise performance data, it seems wiser to have the write
notification be as large as ...

To this end, we need to avoid writing into the notification channel once
the stack has been clo...

We can perform all the operations we need inline.

In particular, make sure we have always seen samples when we
expected to be samples (be them spe...

It was a bit difficult to implement deep packet inspection by
wrapping a network interface, so l...

While that mutation was safe, it is better to not even try.

We can easily make a shallow copy a...

We're not quite there in terms of DPI though. I think we
probably need to redesign it a bit for ...

There is no need to create goroutines in the router when
we can instead let the Link's writer do...

It seems there are some serious issues with the router and the
best way for me to investigate is...

This change allows us to simplify the calibrate command.

The general idea here is that we now have a mechanism for wrapping NICs
when creating a link. Th...

While there, improve documentation.

Let's introduce some library specific errors instead.

When I wrote the prototype inside the OONI codebase, I thought
a periodic counter would be fine ...

The fact that previously we had a method called NIC() that
returned the NIC would have made PCAP...