README_REPOSITORY

vscode: further adjusted default settings and wordPattern for consistent completion popups;
for some reason wordPattern must be set to match (almost) everything, otherwise completions only pop up every other character, and then we must disable wordBasedSuggestions or it will suggest whole lines out of the document;

Important notes on Mercurial repository access for Isabelle
===========================================================

Quick start in 30min
--------------------

1. Platform prerequisites: ensure that "curl" and "hg" (Mercurial) are installed

  (a) Linux: e.g. "sudo apt install curl mercurial

  (b) macOS: e.g. "brew install mercurial" or download from https://www.mercurial-scm.org

  (c) Windows: use Cygwin64 with packages "curl" and "mercurial" (via Cygwin setup-x86_64.exe)

2. Initial repository clone (bash shell commands):

    hg clone https://isabelle.in.tum.de/repos/isabelle
    isabelle/Admin/init

3. Switch repository to particular version (bash shell commands):

    #latest official release
    isabelle/Admin/init -R

    #latest version from repository server
    isabelle/Admin/init -u

    #latest version from local history
    isabelle/Admin/init -u -L

    #explicit changeset id or tag (e.g. "Isabelle2021")
    isabelle/Admin/init -r 479e9b17090e

4. Run application:

    #start Prover IDE and let it build session image
    isabelle/bin/isabelle jedit -l HOL

    #alternative: build session image separately
    isabelle/bin/isabelle build -b HOL

5. Build documentation (bash shell commands):

    isabelle/bin/isabelle build_doc -a

    isabelle/bin/isabelle doc system

6. Find historical evidence about good or bad behaviour
   (e.g. in Isabelle/jEdit):

    cd isabelle
    hg bisect --reset
    Admin/init -R   #alternative: "hg up -r REV"

    Admin/init && bin/isabelle jedit
    hg bisect --good

    Admin/init && bin/isabelle jedit
    hg bisect --bad

    ...


Introduction
------------

Mercurial https://www.mercurial-scm.org belongs to source code
management systems that follow the so-called paradigm of "distributed
version control".  This means plain versioning without the legacy of
SVN and the extra complexity of GIT.  See also
https://www.mercurial-scm.org/learn

Mercurial offers some flexibility in organizing the flow of changes,
both between individual developers and designated pull/push areas that
are shared with others.  This additional freedom demands additional
responsibility to maintain a certain development process that fits to
a particular project.

Regular Mercurial operations are strictly monotonic, where changeset
transactions are only added, but never deleted or mutated.  There are
special tools to manipulate repositories via non-monotonic actions
(such as "rollback" or "strip"), but recovering from gross mistakes
that have escaped into the public can be hard and embarrassing.  It is
much easier to inspect and amend changesets routinely before pushing.

The effect of the critical "pull" / "push" operations can be tested in
a dry run via "incoming" / "outgoing".  The "fetch" extension includes
useful sanity checks beyond raw "pull" / "update" / "merge", although
it has lost popularity in recent years.  Most other operations are
local to the user's repository clone.  This gives some freedom for
experimentation without affecting anybody else.

Mercurial provides nice web presentation of incoming changes with a
digest of log entries; this also includes RSS/Atom news feeds.  There
are add-on history browsers such as "hg view" and TortoiseHg.  Unlike
the default web view, some of these tools help to inspect the semantic
content of non-trivial merge nodes.


Initial configuration
---------------------

The main Isabelle repository can be cloned like this:

  hg clone https://isabelle.in.tum.de/repos/isabelle

This will create a local directory "isabelle", unless an alternative
name is specified.  The full repository meta-data and history of
changes is in isabelle/.hg; local configuration for this clone can be
added to isabelle/.hg/hgrc, but note that hgrc files are never copied
by another clone operation.


There is also $HOME/.hgrc for per-user Mercurial configuration.  The
initial configuration requires at least an entry to identify yourself
as follows:

  [ui]
  username = ABC

Isabelle contributors are free to choose either a short "login name"
(for accounts at TU Munich) or a "full name" -- with or without mail
address.  It is important to stick to this choice once and for all.
The machine default that Mercurial produces for unspecified
[ui]username is not appropriate.

Such configuration can be given in $HOME/.hgrc (for each home
directory on each machine) or in .hg/hgrc (for each repository clone).


Here are some further examples for hgrc.  This is how to provide
default options for common commands:

  [defaults]
  log = -l 10

This is how to configure some extension, which is called "graphlog"
and provides the "glog" command:

  [extensions]
  hgext.graphlog =


Shared pull/push access
-----------------------

The entry point https://isabelle.in.tum.de/repos/isabelle is world
readable, both via plain web browsing and the hg client as described
above.  Anybody can produce a clone, change it locally, and then use
regular mechanisms of Mercurial to report changes upstream, say via
mail to someone with write access to that file space.  It is also
quite easy to publish changed clones again on the web, using the
ad-hoc command "hg serve -v", or the hgweb.cgi or hgwebdir.cgi scripts
that are included in the Mercurial distribution, and send a "pull
request" to someone else.

The downstream/upstream mode of operation is quite common in the
distributed version control community, and works well for occasional
changes produced by anybody out there.  Of course, upstream
maintainers need to review and moderate changes being proposed, before
pushing anything onto the official Isabelle repository at TUM.  Direct
pushes are also reviewed routinely in a post-hoc fashion; everybody
hooked on the main repository is called to keep an eye on incoming
changes.  In any case, changesets need to be understandable, both at
the time of writing and many years later.

Push access to the Isabelle repository requires an account at TUM,
with properly configured ssh to isabelle-server.in.tum.de.  You also
need to be a member of the "isabelle" Unix group.

Sharing a locally modified clone then works as follows, using your
user name instead of "wenzelm":

  hg out ssh://[e-mail]//home/isabelle-repository/repos/isabelle

In fact, the "out" or "outgoing" command performs only a dry run: it
displays the changesets that would get published.  An actual "push",
with a lasting effect on the Isabelle repository, works like this:

  hg push ssh://[e-mail]//home/isabelle-repository/repos/isabelle


Default paths for push and pull can be configured in
isabelle/.hg/hgrc, for example:

  [paths]
  default = ssh://[e-mail]//home/isabelle-repository/repos/isabelle

Now "hg pull" or "hg push" will use that shared file space, unless a
different URL is specified explicitly.

When cloning a repository, the default path is set to the initial
source URL.  So we could have cloned via that ssh URL in the first
place, to get exactly to the same point:

  hg clone ssh://[e-mail]//home/isabelle-repository/repos/isabelle


Simple merges
-------------

The main idea of Mercurial is to let individual users produce
independent branches of development first, but merge with others
frequently.  The basic hg merge operation is more general than
required for the mode of operation with a shared pull/push area.  The
"fetch" extension accommodates this case nicely, automating trivial
merges and requiring manual intervention for actual conflicts only.

The fetch extension can be configured via $HOME/.hgrc like this:

  [extensions]
  hgext.fetch =

  [defaults]
  fetch = -m "merged"

To keep merges semantically trivial and prevent genuine merge
conflicts or lost updates, it is essential to observe to following
general discipline wrt. the public Isabelle push area:

  * Before editing, pull or fetch the latest version.

  * Accumulate private commits for a maximum of 1-3 days.

  * Start publishing again by pull or fetch, which normally produces
    local merges.

  * Test the merged result, e.g. like this:

      isabelle build -a

  * Push back in real time.

Piling private changes and public merges longer than 0.5-2 hours is
apt to produce some mess when pushing eventually!

The pull-test-push cycle should not be repeated too fast, to avoid
delaying others from doing the same concurrently.


Content discipline
------------------

The following principles should be kept in mind when producing
changesets that are meant to be published at some point.

  * The author of changes needs to be properly identified, using
    [ui]username configuration as described above.

    While Mercurial also provides means for signed changesets, we want
    to keep things simple and trust that users specify their identity
    correctly (and uniquely).

  * The history of sources is an integral part of the sources
    themselves.  This means that private experiments and branches
    should not be published by accident.  Named branches are not
    allowed on the public version.  Note that old SVN-style branching
    is replaced by regular repository clones in Mercurial.

    Exchanging local experiments with some other users can be done
    directly on peer-to-peer basis, without affecting the central
    pull/push area.

  * Log messages are an integral part of the history of sources.
    Other people will have to inspect them routinely, to understand
    why things have been done in a certain way at some point.

    Authors of log entries should be aware that published changes are
    actually read.  The main point is not to announce novelties, but
    to describe faithfully what has been done, and provide some clues
    about the motivation behind it.  The main recipient is someone who
    needs to understand the change in the distant future to isolate
    problems.  Sometimes it is helpful to reference past changes
    formally via changeset ids in the log entry.

  * The standard changelog entry format of the Isabelle repository
    admits several (somehow related) items to be rolled into one
    changeset.  Each item is then described on a separate line that
    may become longer as screen line and is terminated by punctuation.
    These lines are roughly ordered by importance.

    This format conforms to established Isabelle tradition.  In
    contrast, the default of Mercurial prefers a single headline
    followed by a long body text.  The reason for that is a somewhat
    different development model of typical "distributed" projects,
    where external changes pass through a hierarchy of reviewers and
    maintainers, until they end up in some authoritative repository.
    Isabelle changesets can be more spontaneous, growing from the
    bottom-up.

    The web style of https://isabelle.in.tum.de/repos/isabelle
    accommodates the Isabelle changelog format.  Note that multiple
    lines will sometimes display as a single paragraph in HTML, so
    some terminating punctuation is required.  Do not squeeze multiple
    items on the same line in the original text!


Testing of changes (before push)
--------------------------------

The integrity of the standard pull/push area of Isabelle needs the be
preserved at all time.  This means that a complete test with default
configuration needs to be finished successfully before push.  If the
preparation of the push involves a pull/merge phase, its result needs
to covered by the test as well.

There are several possibilities to perform the test, e.g. using the
Isabelle testboard at TUM.  In contrast, the subsequent command-line
examples are for tests on the local machine:

  ./bin/isabelle build -a  #regular test

  ./bin/isabelle build -a -o document=pdf  #test with document preparation (optional)

  ./bin/isabelle build -a -j2 -o threads=4  #test on multiple cores (example)

See also the chapter on Isabelle sessions and build management in the
"system" manual. The build option -S is particularly useful for quick
tests of individual commits, e.g. for each step of a longer chain of
changes, but the final push always requires a full test as above!

Note that implicit dependencies on Isabelle components are specified
via catalog files in $ISABELLE_HOME/Admin/components/ as part of the
Mercurial history.  This allows to bisect over a range of Isabelle
versions while references to the contributing components change
accordingly.  Recall that "isabelle components -a" updates the local
component store monotonically according to that information, and thus
resolves missing components.