Load (external) source before actually renaming / filing out classes / extensions Making sure the source is loaded must be done before any changes to the working copy are made. Imagine one moves class A from package 'p' to subpackage 'p/s'. If first 'p' is filed out and then 'p:s' is filed out then class A sources would be gone because fileout of 'p' would remove the file. This commit does not solve all the issues but should help in this case.

`SCMAbstractCommitDialog`: specify text color on header pane This must be done since we specify backgroung color (as white) and therefore cannot depend that theme-defined text color would be good and contrasting.

Fix for truncated HGChangesetId at push parsing errors - fixes HGStXTests>>tests_commit_05b for Mercurial 5.8

Fix (bad) source corruption when committing private classes This commit fixes a really bad bug causing source corruption when committing private classes *and* when using original repository for commits ( `hgUseOriginalRepositories: true`). This was becase private classes's source was not made local prior opening the enclosing class's container for writing! Sigh.

Converting the windows style path (with '\') into unix one ('/'). The filenames at HGCommandParser>>parseCommandLocate changed the path style on Windows.

Fix `HGContribution class >> initialize` Recent `Timestamp` parser is more strict and does parse when only date is given. Makes sense.

Add (experimental) option to use original working copy for commits ...rather than (possibly shared) temporary one. Use of temporary one makes a lot of things more complex and error-prone (such as, amending commits and handling bookmarks). Moreover, it is confusing for most users and hard to understand what has happened when something goes wrong. This should make things simpler and more like things work in non-Smalltalk world.

Rework and fix HGSourceCodeManager >> #revisionLogOf:...directory:module:` This commit changes the logic in two ways: 1. #newestRevision is now the newest revision in the branch that *contains* given file (not necesarily modidfes it). If there are multiple heads in that branch, pretty much random one is returned. This changes old behavior and therefore this commit updates tests. 2. If a specific single revision is requested, i.e., both from and to revisions are the same, revision log with that single revision is returned no matter whether it modifies the file or even contains that file at all. This is essentially a workaround to fix issue #305. Moreover, this commit simplifies the code a lot by delegating all the changeset searching and filtering to mercurial using revset expressions. See https://swing.fit.cvut.cz/projects/stx-jv/ticket/305#comment:3

Fix issue with bookmark not advancing upon commit ...when original repository is used. This was caused by forcefull updating of working copy to a revision, which deactivates active bookmark.

Add tests checking that bookmark is moved upon commit