Skip to content

Release 2.3.2 - #409

Merged
JeanMainguy merged 37 commits into
masterfrom
release/2.3.2
Sep 28, 2026
Merged

JeanMainguy merged 37 commits into
masterfrom
release/2.3.2

Conversation

@github-actions

@github-actions github-actions Bot commented Sep 28, 2026 •

Copy link
Copy Markdown
Contributor

Release 2.3.2

This PR prepares the PPanGGOLiN 2.3.2 release.

Included PRs

Release changes

  • Update VERSION to 2.3.2
  • Update expected files containing the PPanGGOLiN version

JeanMainguy and others added 30 commits August 13, 2026 11:47
Add generated command-line reference documentation
write_gexf_header is the only GEXF writer that builds its header by hand,
and it declared xmlns="https://www.gexf.net/1.2draft". The 1.2draft
schema declares targetNamespace="http://www.gexf.net/1.2draft", so the
https form matches nothing.

Gephi's importer dispatches on element local names and ignores the
namespace URI, which is why these files still open there. A
namespace-aware reader does not: networkx raises
NetworkXError("No <graph> element in GEXF file.") on both
pangenomeGraph.gexf and pangenomeGraph_light.gexf, so neither can be
loaded in Python without patching the file first.

Every other GEXF output (spot graphs, rgp_cluster, context, draw_spot)
goes through nx.write_gexf and already emits the correct namespace, so
this also removes an inconsistency between our own files.

Introduced in 00978da and present in every release from 1.2.46 to
2.3.1. That commit also added the "# TODO update link" comment on the
line it changed, so the comment goes with the fix.
The functional test asserted only that the gexf output exists and is
larger than zero bytes, which a file with a malformed header satisfies,
so it did not catch the namespace bug fixed in the previous commit.
Before this test suite landed in 2.2.6, nothing checked the gexf output
at all.

Read the file back with networkx, already a dependency, and require a
non-empty graph whose nodes still carry their viz attributes.

The viz check is not redundant: fixing only the default namespace and
leaving xmlns:viz on https still parses, but silently drops every
partition colour and size, and a node-count assertion alone would pass.
networkx reads the compressed .gexf.gz directly, so the existing
parametrisation needs no change.
`ppanggolin info` opens the file in "r+", and the metric check in
`metrics` opens it in "a", but neither writes. info calls read_status,
read_info, read_parameters and read_metadata_status; the metric check
only inspects h5f.root.info._v_attrs. Every reader in readBinaries.py
already opens in "r".

Both therefore fail on a pangenome the user cannot write:

    PermissionError: file ``pangenome.h5`` exists but it can not be written

That rules out inspecting a pangenome on read-only media — a shared
reference collection, a read-only mount, a container volume — and it
rules out `chmod 444` as a guard against a command rewriting a file in
place. `metrics --genome_fluidity` is affected even when the fluidity is
already stored and only needs printing, because
check_already_computed_metric runs before anything decides whether to
compute, and is not guarded by any flag.

Both modes are leftovers. info needed write access only for the
Partitionned -> Partitioned repair: 78390bb (2022) raised the mode to
"r+" alongside that repair, which reopened the file in "a" to do its own
write; the repair later moved out of print_info into fix_partitioned(),
lost its last caller in 5b2942a, and the function was deleted in
82330ab. The metric check has opened "a" ever since check_metric was
introduced in 3795e50 (2022); 079bf87 (2023) dropped the --force guard
in front of it, so it now runs on every invocation.

Output is unchanged: `ppanggolin info` on the same pangenome produces
byte-identical text before and after, and the file's md5 is untouched.
Copy the fixture, chmod 444, then run both commands. Each mode change is
independently necessary: reverting both fails on info before metrics is
reached, and reverting only the metric check still fails, on metrics.

The chmod is only effective for an unprivileged user, so the test skips
itself when write access survives it — under root the commands would
succeed unfixed and the test would silently prove nothing. PyTables gates
its write modes on exactly the os.access() call the guard uses.
#391)

* rename contributing doc so it is detected by github

* mv contributing doc so it is detected by github

* fix ref of  CONTIRBUTING doc

* add how to contribute in the readme directly with a link to the doc

* Update RGP clustering documentation to clarify default output directory and file naming

* Clarify formatting and fix typos in RGP clustering documentation

* Clarify annotation file usage and input requirements in pangenomeAnnotation and projection documentation

* Fix typos and improve table formatting in projection documentation

* Fix incorrect link in projection documentation

* Add "What's new in PPanGGOLiN v2" documentation

- Summarize major changes and new features introduced in version 2.0.0, including new commands, reorganized commands, new output formats, usability improvements, and performance enhancements.
- Update index to include the new page.

* Update "What's New in PPanGGOLiN v2" documentation

- Expand details on version boundary and compatibility changes.
- Add new commands (`context`, `all`, `metrics`) and reorganized command structure.
- Revise formatting for improved readability.

* Fix typos and improve formatting in "What's New in PPanGGOLiN v2" documentation

- Correct minor typos and standardize American English spelling.
- Refine table formatting for consistency and readability.
- Optimize phrasing in usability and performance sections.

* Expand metadata documentation with detailed example for annotating pangenome gene families using Pfam

- Add step-by-step guide for exporting sequences, running annotations, formatting results, and adding metadata.
- Clarify metadata formatting rules and valid identifier requirements.
- Improve readability with example tables and inline notes.

* Fix typos, improve table formatting, and standardize phrasing in metadata documentation

* Removed unnecessary line

Co-authored-by: Jean Mainguy <jean.mainguy@outlook.fr>

* Remove overclaimed line

Co-authored-by: Jean Mainguy <jean.mainguy@outlook.fr>

* Remove citation section to wait for actual publication

Co-authored-by: Jean Mainguy <jean.mainguy@outlook.fr>

* Remove citation section to wait for actual publication

Co-authored-by: Jean Mainguy <jean.mainguy@outlook.fr>

* Fix grammar and format

Co-authored-by: Jean Mainguy <jean.mainguy@outlook.fr>

* Update introduction in whatsNewInV2.md

Expanded the introduction to include references to panRGP and panModule, providing a broader context for users transitioning from version 1.

---------

Co-authored-by: Jean Mainguy <jean.mainguy@outlook.fr>
* fix gexf namespace so the pangenome graph can be parsed

write_gexf_header is the only GEXF writer that builds its header by hand,
and it declared xmlns="https://www.gexf.net/1.2draft". The 1.2draft
schema declares targetNamespace="http://www.gexf.net/1.2draft", so the
https form matches nothing.

Gephi's importer dispatches on element local names and ignores the
namespace URI, which is why these files still open there. A
namespace-aware reader does not: networkx raises
NetworkXError("No <graph> element in GEXF file.") on both
pangenomeGraph.gexf and pangenomeGraph_light.gexf, so neither can be
loaded in Python without patching the file first.

Every other GEXF output (spot graphs, rgp_cluster, context, draw_spot)
goes through nx.write_gexf and already emits the correct namespace, so
this also removes an inconsistency between our own files.

Introduced in 00978da and present in every release from 1.2.46 to
2.3.1. That commit also added the "# TODO update link" comment on the
line it changed, so the comment goes with the fix.

* check that written gexf files can be parsed back

The functional test asserted only that the gexf output exists and is
larger than zero bytes, which a file with a malformed header satisfies,
so it did not catch the namespace bug fixed in the previous commit.
Before this test suite landed in 2.2.6, nothing checked the gexf output
at all.

Read the file back with networkx, already a dependency, and require a
non-empty graph whose nodes still carry their viz attributes.

The viz check is not redundant: fixing only the default namespace and
leaving xmlns:viz on https still parses, but silently drops every
partition colour and size, and a node-count assertion alone would pass.
networkx reads the compressed .gexf.gz directly, so the existing
parametrisation needs no change.
Open the pangenome read-only when only reading it
Add example input files to align and projection documentation
* add default in help text

* Support Python 3.10–3.14 and drop Python 3.9

* only test 3.14 on ubuntu

* exclude python 3.14 because of bioconda

* update min numpy version
* adding gb_io to parse gbff : first steps

* gb_io unit test fixing

* add gb-io to env

* black formatter

* black formatter

* change gbff opening + removing old functions

* Update Python version in CI workflow

---------
* fix: replace aragorn subprocess with pyaragorn package (#401)

* fix: remove pycache

* remove .gitignore

---------

Co-authored-by: fathirramadhan-web <fathirramadhan.web@gmail.com>
Co-authored-by: JeanMainguy <jean.mainguy@outlook.fr>
JeanMainguy and others added 7 commits September 23, 2026 10:26
…d FASTA inputs (#405)

* improve input validation and add clearer message

* apply black
…#407)

* update in the command line ref doc automatically in CI

* docs: regenerate command reference

* trigger build doc only when doc change and trigger command-ref check only when cli change

---------

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
* order genomes deterministically

* seed K estimation

* fix naming

* fix naming again
* rgp_cluster optims draft

* parse and compute is_border and whole_contig for rgp proxy

* add pyroaring to deps

* not load of contigs with no RGP and read twice gentable to not store it in mem

* add timer on _construct_regions

* add ressource peak mem monitoring

* make rgp output the same with old implementation

* add metadata management in new implementation

* prefilter rgp pairs to not compute GRR on all possible combinations and make clustering determinist

* implement ignore_incomplete_rgp in new implem

* add progress bar when computing metrics and fix no_identical_rgp_merging param

* add check on table existence to fail nicely if missing

* apply black ../ppanggolin/utils.py

* remove old rgp cluster implementation

* fix black in utils

* add warning when pangenome has a RGP table but has 0 RGPs in it (small pangenomes with almost only persistent families)

* make graph format graphml only by default

* update unit test for rgp clustering

* apply black

* remove metrics list attribtue of the RGPClustering  because it is not used anywhere

* fix test according new default graph format output

* apply black

* keep only genes in RGP when geting them from h5. and keep self._rgp_to_genes only when with_metadata is set

* add slots=True for dataclass

* Revert "add slots=True for dataclass"

This reverts commit 53baf5d.

* sort graph attributes

* add docstring to every fct and methods

* add slots=True for dataclass

* fix merge conflict

---------

Co-authored-by: Teo Lemane <teo.lemane@gmail.com>
@JeanMainguy
JeanMainguy merged commit 8804f27 into master Sep 28, 2026
10 checks passed
@JeanMainguy
JeanMainguy deleted the release/2.3.2 branch September 28, 2026 13:05
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

7 participants