Differences between revisions 80 and 81
Revision 80 as of 2011-01-19 09:49:53
Size: 22105
Comment:
Revision 81 as of 2011-01-19 12:12:59
Size: 22227
Comment:
Deletions are marked like this. Additions are marked like this.
Line 213: Line 213:

# set the shortusername variable to the letters that define your signature on the patches (e.g. 'sc' for Simon Cussonet)
Line 214: Line 217:
pre-qrefresh = (shortusername='azerty' ; if [[ "$(sage -hg qtop | grep -c "\-$shortusername.patch$")" != "1" ]] ; then echo -e "\\033[01;05;31m* * * * * * * THIS IS NOT YOUR PATCH * * * * * * *\\033[0;0m" ; fi ; echo "Are you sure you want to refresh the following changes:"; sage -hg status; echo -n "into the patch: "; sage -hg qtop; read -p "(y/n)" answer; test "$answer" = "y") pre-qrefresh = (shortusername='sc' ; if [[ "$(sage -hg qtop | grep -c "\-$shortusername.patch$")" != "1" ]] ; then echo -e "\\033[01;05;31m* * * * * * * THIS IS NOT YOUR PATCH * * * * * * *\\033[0;0m" ; fi ; echo "Are you sure you want to refresh the following changes:"; sage -hg status; echo -n "into the patch: "; sage -hg qtop; read -p "(y/n)" answer; test "$answer" = "y")

The Sage-Combinat patch server: step by step instructions

This page is meant as a step-by-step tutorial on using the Sage-Combinat patch server, from basic installation to contributing new patches:

For technical background on the patch server, see http:/combinat/Mercurial.

To do: add a link to a short write up about the rationale for using a patch server.

1. Installation and basic usage

1.1. Prerequisites

The instructions below assume that:

  • You need to have a very recent version of Sage already installed (say in /opt/sage) which can be started by typing sage at the command line

    • (it is recommended to use the most recent stable version, and in particular >= 4.3.1 as of 2010, march the 20)

  • The user has write access to the Sage installation tree

  • The user has access to: http://combinat.sagemath.org/patches/

Notes:

  • Pushing patches to the server requires you to authenticate with your trac server username and password. Create them as indicated on the trac server page. To avoid having to type your username and password each time you push patches to the patch server, you can include the following lines in your .hgrc configuration file (see more about this file in the "Mercurial configuration" section further down):

    [auth]
    combinat_patches.prefix = http://combinat.sagemath.org/patches/
    combinat_patches.username = trac_login
    combinat_patches.password = trac_password
  • The patch management is based on the version control system Mercurial, with a support script sage -combinat for ease of use. Online help on this script is accessible through:

    sage -combinat --help

1.2. Downloading and installing the Sage-combinat patches

sage -combinat install

1.3. Updating the Sage-Combinat patches

To update the Sage-Combinat patches to the latest version, you can run:

sage -combinat update

It is possible to simultaneously upgrade Sage to the latest version:

sage -combinat upgrade

However, we do recommend installing instead a fresh new version of Sage, and then reinstalling the patches on top of it with sage -combinat install.

1.4. Uninstalling the Sage-combinat patches

sage -b main

Now, you may destroy .../sage/devel/sage-combinat/ (after checking that you do not have local changes!)

2. Looking and selecting the patches

2.1. The Sage-combinat stack of patches

Sage-combinat is a collection of experimental patches (i.e. extensions) on top of Sage. Each patch describes a relatively atomic modification which may span several files; it may fix a bug, implement a new feature, improve some documentation. Then main information contained in a patch is a diff file which describes the difference between two files. Here is an example of a diff file.

diff -r 01fabd9b3951 sage/combinat/subword.py
--- a/sage/combinat/subword.py
+++ b/sage/combinat/subword.py
@@ -34,7 +34,8 @@

 def Subwords(w, k=None):
     """
-    Returns the combinatorial class of subwords of w.
+    Returns the combinatorial class of subwords of w. The word w can be given
+    by either a string or a list.

     If k is specified, then it returns the combinatorial class
     of subwords of w of length k.

It reads as follows: the first four lines tells that this patch modifies the file sage/combinat/subword.py. The line starting with - is replaced by the two lines starting with +. To be able to apply the modifications safely at the right place, some context information is also stored (ie: some more unmodified lines and the position of the modification into the file).

The patches are organized as a (totally ordered) stack, each being applied on top of the previous one. The most stable are at the bottom, while the most experimental ones are on top. Let's look at a typical stack of patches:

sage -hg qseries

It will display something like:

sage-3.1.3.patch
sage-3.2.patch
...
lazy_attributes-4371-nt.patch
copy_on_write_fh.patch
discrete_function-nt.patch
crystals_alcove_path_model_bj.patch
words_new_fcts_sl.patch

sage-3.1.3 is at the bottom of the stack, and is applied first, while words_new_fcts_sl.patch is applied last. The patch sage-3.2.patch contains all the sage-combinat patches that have readilly been integrated in sage 3.2 (see "using sage-combinat with older versions of sage"). The patch lazy_attributes-4371-nt.patch is about lazy attributes, related to ticket number 4371 on Sage trac, and is owned by NicolasThiéry (nt).

In practice, the stack of patches is managed as a Mercurial queue. The command sage -hg calls the version of Mercurial that comes with Sage. You may instead simply use hg if Mercurial is installed natively on your system.

2.2. Top, applied and unapplied patches

It is possible to move up and down the stack, applying only the patches up to a given one.

sage -hg qpush             # Apply the first patch in the series which is not currently applied
sage -hg qpop              # Unapply the most recently applied patch
sage -hg qpush -a          # Apply all the patches
sage -hg qpop -a           # Unapply all the patches

If you get confused, the following can tell you which patches are applied or not:

sage -hg qtop              # Top applied patch
sage -hg qapplied          # Currently applied patches
sage -hg qunapplied        # Currently unapplied patches

Note that after moving around the stack of patches, it is necessary to rebuild sage before using it :

sage -b
sage

2.3. Looking inside patches

The description of a patch can be retrieved with:

sage -hg qheader                      # top patch
sage -ht qheader bla.patch

The content of the current top patch can be retrieved with:

sage -hg qdiff

while the files modified in the top patch can be seen with:

sage -hg qstatus

Warning:  sage -hg qstatus  will return irrelevant information if there is no patch currently on top of the stack.

2.4. Using the sage-combinat patches with an older version of sage

For the convenience of the user, it is usually possible to use the sage-combinat patches with older versions of sage. The intent is only to temporarily support one or two older versions of sage (that is about one month old). Typical use case: a developer urgently needs the latest version of a patch for a software demonstration at a conference, but can't instantly upgrade because of a slow internet connection. There is no guarantee whatsoever; on occasion we do not support this when this causes technical difficulties.

This is the purpose of the patches like sage-3.0.2.patch. This particular patch contains all the sage-combinat patches that have been integrated into Sage 3.0.2, and is only applied if Sage's version is strictly less than 3.0.2. This is achieved through guards (see the next section), and is taken care of automatically by sage -combinat, and in particular by:

sage -combinat qselect

2.5. Selecting guarded patches

Some patches may be guarded since they are experimental, not yet finished, or should be only applied for certain versions of sage. Guarded patches are not applied unless explicitly chosen. For example if one would like to apply patches labeled [+experimental] one can use the following steps:

sage -hg qselect experimental
sage -combinat qselect

Then reapply the appropriate patches, typically with:

sage -hg qpop -a
sage -hg qpush -a

To disregard all guarded patches one uses instead:

sage -hg qselect -n
sage -combinat qselect

3. Creating and contributing patches

3.1. Mercurial configuration

Some configuration is required for write access to the Mercurial server. Open (or create, if it does not exist yet) your Mercurial configuration file (~/.hgrc in your home directory), insert the following, and edit the username, and the shortusername (usually the 2 initials of your first-name/last-name):

[ui]
username = Simon Cussonnet <Simon.Cussonnet at lycee-technique.thorel.eure.fr>

[auth]
combinat_patches.prefix = http://combinat.sagemath.org/patches/
combinat_patches.username = <your username on Sage's trac server>
combinat_patches.password = <your password on Sage's trac server>

[extensions]
hgext.mq =
hgext.highlight=
hgext.record=
extdiff =
color=

[alias]
qstatus = status --rev -2:.


# set the shortusername variable to the letters that define your signature on the patches (e.g. 'sc' for Simon Cussonet)

[hooks]
pre-qrefresh = (shortusername='sc' ; if [[ "$(sage -hg qtop | grep -c "\-$shortusername.patch$")" != "1" ]] ; then echo -e "\\033[01;05;31m* * * * * * *  THIS IS NOT YOUR PATCH  * * * * * * *\\033[0;0m" ; fi ; echo "Are you sure you want to refresh the following changes:"; sage -hg status; echo -n "into the patch: "; sage -hg qtop; read -p "(y/n)" answer; test "$answer" = "y")
pre-import = (echo "I'm pretty sure that you don't want to import a changeset !"; echo "Do you mean 'hg qimport <patch>' ?"; exit 1)

[extdiff]
cmd.interdiff = hg-interdiff

[diff]
nodates=1
showfunc=1
git=1

3.2. Patch naming convention

Each patch should be of the form:

<trac ref if any>-<theme>_<feature/fix>_<rev if any>-<owner's shortusername or final or closed>.patch

Examples:

trac_9557-fundamental_domains-vd.patch
sturmian_words_classes-abm.patch
trac_5991_dynamic_class-nt.patch
trac_7236_partitions_tableaux_cells_cleanup-fh.patch
trac_10063-determinant_not_computed_in_some_rings_bugfix_attempt_4-tm.patch
trac_7251_integer_parent-nt.patch
trac_5038_paths-sl.patch
graphs_paths_and_cycles_enumeration-abm.patch
free-modules-final.patch
dyck-words-closed.patch

A series of patches like the free_modules one above is intended to be progressively folded together into a single patch free_modules.patch before submission to sage (see patch folding below). The name free_modules_final indicates that the owner of this patch will no longer edit it, and it is ready to be posted to trac. The name *_closed.patch indicates that this patch has been posted to trac and the corresponding ticket has been closed.

3.3. Patch description

Each patch starts with a description header which is preserved upon hg export. This is the first thing that will be seen by someone opening the patch. Moreover, the first line of this description will end up into the history of Sage (see the output of sage -hg log). So it should be as meaningful as possible, i.e. contain the ticket number associated to the patch and a small description.

Typically, this description will match with the summary + description of the (upcoming) trac ticket:

#5431: Free modules: implements blah blah

This patch implements ...

This description can be edited at any point using hg qrefresh -e, so feel free to update it on a regular basis to include progress information (what has readily been implemented, what remains to be done, etc:

#1234: Permutations: implementation of bruhat order

Done:
* Added methods succ_bruhat, pred_bruhat, ...
Todo:
* Add method bruhat_order, ...

This is usually at this point that one has the best view of this description.

3.4. Creating a new patch (qnew)

The first step is to create a new patch. It is easier to create it before doing any modifications to the sage sources. Never modify a patch that you do not own.

sage -hg qnew my_improvement_ab.patch -m "#N: description of the patch"

where ab are your initials and N is the associated sage trac ticket number if it exists. For a larger description you may want to use the -e option to edit the description with your favorite editor.

The new patch is created on top of the most recently applied patch. You may use qpush and qpop first to choose where your patch is created. If you accidentally made changes before creating a new patch, the command

sage -hg qnew -f my_improvement_ab.patch -m "#N: description of the patch"

will create the new patch including the current 'orphaned' changes into it.

3.5. Editing the Sage sources

It is recommended to double check that the current top patch is yours and is the one you want to add your modifications to (see qtop and qpop/qpush). using qpop and qpush become tricky to use once you started modifications.

Now you can edit the Sage sources files to your taste in $SAGE_ROOT/devel/sage-combinat/sage/. At any time, you can review your modifications since the creation of the patch (or since the last qrefresh, see below) by doing:

sage -hg diff           #complete modifications since last qrefresh
sage -hg status         #list the modified files since last qrefresh

If you added a new source file, you must declare it using the command:

sage -hg add <filename>

3.6. Refreshing your patch (qrefresh)

Currently, the modifications are still not part of the patch as may be seen with the commands

sage -hg diff    # Completely lists the modifications
sage -hg status  # Lists only the affected files

Note that sage -hg status gives the modifications which are not part of the current patch, while sage -hg qstatus gives the modifications which are part of the current patch (and similarly for diff and qdiff).

Use

sage -hg qrefresh

to put the actual modifications in the current top patch.

Tip: on a regular basis, use also:

sage -hg qrefresh -e

to update the description of the patch, as this usually is the point in time where one has the best view of what it should be. It can also be the opportunity to add the sage trac ticket number to the description if it is created.

You can check that the modifications have been included in the patch with:

sage -hg qdiff
sage -hg qstatus

Also, they should not appear anymore in:

sage -hg diff
sage -hg status

After qrefresh, you can use qpop and qpush again and modify the same patch or some other patch you already created.

3.7. Removing a patch

In case you want to discard your patch, you may use:

sage -hg qremove my_improvement_AB.patch

You may use

sage -hg qseries

to confirm that the patch is removed.

3.8. Pushing patches to the sage-combinat server

Go to the sage-combinat directory, and make sure that there is no local changes:

cd $SAGE_ROOT/devel/sage-combinat
sage -hg status

Otherwise discard them (sage -hg revert) or save them in your favorite patch (sage -hg qrefresh).

Pull the latest version of the patches from the server, and make sure that everything is working fine after applying all patches:

sage -combinat update -f
sage -hg qpush -a
sage -br  # Make sure this works
sage -t filenames # If you believe your tests should pass, check them.

The -f is to tell sage -combinat to run even if your patch directory is modified, which it probably is.

When pulling from the server, your changes will be merged with those from the server. You may get conflicts in the series file, or patches that do not apply anymore. Fix the patches that you own, and get in touch with the owners of the other broken patches.

When everything is ready, double check that you are up to date:

sage -combinat update -f

and then:

cd $SAGE_ROOT/devel/sage-combinat/.hg/patches
sage -hg commit
sage -hg push

In case you have been unlucky, and there has been a change on the patch server between your last pull and your commit, push will fail with an error like "abort: push creates new remote heads! did you forget to merge?". You then have to merge your local modifications with the remote ones:

cd .hg/patches
sage -hg pull -u
sage -hg merge

If the merge goes through automatically, just commit with "Merge" as message, and push. Otherwise, read the section Merging conflicts and feel free to ask for help!

3.9. In the case of Merging conflicts

In the case of merging conflicts, you should obtain such a message :

TOBEDONE

First, use the following to list the files needing a manual merge. If you edited your own patches and nobody edited your patches, only the series file should be listed. If other files are listed, you should definitively ask for help.

cd .hg/patches
hg resolve -l #list state of files needing merge (U = unresolve, R = resolved)

Open and edit the series file and merge the modifications yourself. Then, when you are done, mark the series file as resolve by doing

hg resolve -m series

Make sure the series file is now marked as resolved using:

hg resolve -l #list state of files needing merge (U = unresolve, R = resolved)

Then, try to merge again

hg merge

and commit if the merge worked:

hg commit

You may now try to push again to the sage-combinat server as explain above.

3.10. Handling rejection

Sometimes a patch fails to apply and you get, after a

sage -hg qpush

a message like :

applying my_modification.patch
patching file file_a
Hunk #1 FAILED at 0
1 out of 17 hunks FAILED -- saving rejects to file file_a.rej
patch failed, unable to continue (try -v)
patch failed, rejects left in working dir
Errors during apply, please fix and refresh
my_modification.patch

After such an error, Mercurial has made all modification its can to the file file_a and rejects the hunk its could not understand. All modifications inside the patch and not understood by Mercurial are in the file file_a.rej created in the working directory (e.g. at the same place that file_a is). The rejected file is just a part of the initial patch.

At this point you have several solutions, the first is more simple :

  • make by hand the modification in file_a.rej to file_a and refresh the patch my_modification.patch
  • try to reorganize directly the patch my_modification.patch (DANGEROUS)

  • use the Chris Mason's mpatch to try to reorganize the patch (automatization of the preceding solution).

In any case be wise with any modification !

3.11. Exporting Patches for use with trac

When a patch is ready to be applied to Sage, you should first verify that it will apply cleanly. If you want to be able to undo anything you do in this step, you should run sage -hg qcommit  before beginning.

The first step is, from the PATCHES directory, manually edit the series  to move your patch in the queue, so that your patch (or patches) are applied only to patches labelled *_closed. Then (from the source directories) use qpush and qpop to bring the first of your patches to the top of the stack (ie, only one of your patches should be applied). Suppose your patches were called 1.patch and 2.patch, and you currently have 1.patch at the top of the stack, and 2.patch as the next patch to be applied. The command sage -hg qfold 2.patch will incorporate the changes from 2.patch into 1.patch, and then delete 2.patch. You should repeat this process until all patches working on one particular issue have been incorporated.

Double check and update the description of the patch, using hg qrefresh -e and make sure to add the sage trac ticket number to the first line of the description. You may also rename the patch (using sage -hg qrename) by appending trac_#### as a prefix.

Then you should make sure that sage builds, runs, and passes tests, with just this patch applied on top of the patches that have already been closed. Don't forget to also commit and push your changes to the server.

The next step is to export your patch for submission. Running sage -hg export includes extra information in the patch such as the author of the patch.

cd $SAGE_ROOT/devel/sage-combinat
sage -hg export new_feature.patch > /path/to/new_feature.patch

Upload the file new_feature_for_trac.patch to the appropriate ticket on the Sage Trac server, creating the ticket if necessary. Generic procedures for using the trac system for Sage are described here: http://wiki.sagemath.org/TracGuidelines . Here are some further tips for Sage-Combinat tickets:

  • Keep the same patch name in the Sage-Combinat queue and on trac.
  • Assign it to milestone sage-combinat initially. Once it has a positive review it should be moved to the current merge milestone that is active.
  • Add "sage-combinat" to the CC field. That way ticket status changes show up at http://groups.google.com/group/sage-combinat-commits

4. Rebasing the patch queue on a new version of sage

This section is for maintainers of the patch queue. Idealy, switching to a new version of Sage should require no work. But sometimes some patches from the Sage community at large that were integrated in the new version of Sage may conflict with the Sage-Combinat patches. Here is how to rebase on them. In this example, we rebase from sage 3.4 to sage 3.4.1. This requires having access to vanilla sage/devel/sage-main directories for 3.4 and 3.4.1 which we assume to be respectively in /opt/sage/devel/sage-main-3.4 and /opt/sage/devel/sage-main

Start from a fresh sage-main from the previous sage version:

cd /tmp
hg clone /opt/sage/devel/sage-main-3.4 sage-main

Install the sage-combinat patches, and apply them all

cd sage-main/.hg
hg clone http://combinat.sagemath.org/patches patches
cd ..
hg qselect 3_4_1
hg qpush -a

Then we save the queue state, and launch a 3-way merge

hg qsave -e -c
hg pull /opt/sage/devel/sage-main
hg update -C tip
hg qselect -n
hg qpush -m -a