This matters when we're rerunning previous measurements using the cache.

Without this fix, the ...

These changes are based on my trying to make sense of the
several measurements I'm using as test...

1. we need to rewrite the IDs when we pick results up from the cache
because the ID in general a...

This is required to re-run measurements using the cache.

While there add support for asbolute t...

I should have added this code to miniooni long time ago.

We're using the same format of minioon...

This seems one of the missing bits to ensure that, when we're
repeating a measurement, we're not...

We _also_ need to cache dnsping results to make measurements fully
reproducible later in time. W...

This diff introduces changes allowing one to use the websteps command
to fill a predictable, com...

This command line design is more flexible.

If we only use one, we're not able to compare the DNS resolver and
the system resolver to say wh...

This should help me implement my plan of having two local caches
for re-running measurements loc...

1. stop relying on hashing because we're not convinced

2. cleanly split each part of the analys...

This path was a bit too tricky, so I'll try a simpler path that does not
require me to separate ...

Inspect is informing the user we're doing an activity.

Scrutinize is telling the user we're loo...

Otherwise, we may end up missing some messages.

It's very seldom used and so let's just not use it.

This message is an info message associated with 🧐.

The previous code was not obvious and I could not fully explain why
it was working or non-workin...

In greedy mode we stop following redirects as soon as we find
some signs of internet censorship....

If we also use dnsping, then analysis is going to have a hard
time because some TH endpoint meas...

1. ensure that the probe is going to test in its additional round
only new IP addresses returned...

After reasoning a bit more on it, it turns out that
d46101418537113d15c05d7309f4de8fcae28242 is ...

We can reduce the complexity of Int64 by just ensuring its int64 value is
the first element: thi...

With embedding, every time we add a new method, we don't get a compiler
error, which means the n...

We encode emojis inside the loglevel and each consumer chooses
whether it wants messages with em...

We now have a ring buffer where we store recent log messages, which
allows the mobile app to see...

There is still _more_ work to do in this space, like determining the
full list of acceptable red...

This commit replaces model.Logger instances pretty much everywhere with
calls to the logcat pack...

This query collects extra information useful to further
disambiguate possible DNSDiff cases.

If we cut the name we cannot see what was matching. I am also
wondering whether we should invest...

1. we need to serialize headers as a list rather than as a map
because in golang maps have unpre...

The correct referer is the previous URL rather than the location.

We should not pass extra head...

With these information hopefully we're able to say enough on any
given lookup w/o more complicat...

1. collapse probe_initial into the single-step measurement, which
would be annoying to do when g...

We should not exaggerate with the number of IP addresses tested by
the test helper. Let's add on...

For websites with _many_ IP addresses like YouTube, this would
mean testing a very large number ...

This functionality is very useful when reading logs.

I think I had a bunch of false positives caused by the fact that
the test helper was not actuall...

1. cache algorithms were wrong in that they did not prune
duplicate entries but just added new e...

By knowing it, we can emit more useful log messages.

We cannot exclude the possibility that a dns measurement and an
endpoint measurement could have ...

Create a summary of an endpoint that also includes all the
relevant options. So, we're now also ...

While caching by IP address is nice in principle, we are going to
possibly have several records ...

This change requires adding support for extracting the CNAME from
a DNS round trip result. We st...

It's easier to reason on a JSON when fields with their default
value are actually missing from t...

We need to know their values to construct a more precise
representation of endpoints where we'll...

We were missing this field, which was not so good in terms of
heuristics for detecting HTTP diff...

With a mandatory cache it's too difficult to develop in the
common case where I need to test wit...

Store all endpoints by IP address and check for equality using
the current summary (we still nee...

We're now caching on a domain basis. We have improved the rules
to check whether a plan and a me...

When searching for a compatible DNS lookup measurement, we should
first try to find a "same as" ...

It's actually fine to cross compare getaddrinfo and HTTPSSvc as
long as both have gathered some ...

We need to refactor the code until there is a single plan for
each measurement. This is the best...

This diff changes the way in which we create DNSLookupPlans to
produce a single plan per resolve...

These flags allow one to easily specify extra queries to
perform (e.g., HTTPSSvc or NS) as part ...

The idea here is to never expire the cache. This allows us to
test the cache performance. It doe...

This should allow us to retry measurements at a later time w/o
having to really do network I/O.
...

The previous diff was not correct, sadly. So we need a new diff.

Otherwise the cache ends up being owned by root and it does not
work because the unprivileged pr...

We import the codebase used by Go to maintain its package cache, which
has been modified to remo...

The AbstractMeasurer is an interface that replaces a concrete
Measurer with an abstract version ...

Basically, only transmit fields that are different from their
default value. This change makes t...

Too often they do not work or are censored just like normal
queries, so it seems more economic t...

This diff modifies the emitted data format so that:

1. we only emit DNS round trips

2. we emit...

The advantage of using a (obviously fake) transport for the system
resolver is that we can colle...

At the measurex level, you can ask for a NS lookup by using
a flag that extends the performed qu...

If we're measuring http://torproject.org with the default config we'll
end up using just two of ...