fix(composer): consume icewind deps from GitHub mirrors to survive Codeberg outages - #41707

…deberg outages [10.16] (#41816)

fix(composer): consume icewind deps from GitHub mirrors to survive Codeberg outages

Backport of #41707 to 10.16.

owncloud CI intermittently fails because Composer must git-clone
icewind/streams and icewind/smb from codeberg.org (no dist archive on
Packagist), and Codeberg regularly returns HTTP 503/504. Every job that
installs Composer dependencies is affected, including "Make local core"
in downstream app builds.

Add VCS repositories pointing at the GitHub mirrors so Composer fetches a
dist zipball via the GitHub API and never touches Codeberg at install time.

Unlike master, 10.16 stays on icewind/smb v3.7.0: the 3.8.x line requires
php >= 8.2 while this branch pins platform php 7.4. Both packages are
re-locked at the exact refs already in the lock (smb e6904cb, streams
cb2bd3e), which the mirrors carry, so this is a pure source repoint with
no version change. No codeberg.org references remain in composer.lock.

Mirrors are kept in sync weekly by DeepDiver1975/codeberg-mirrors.

Signed-off-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>
Co-authored-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>

…deberg outages [10.16] (#41816)

fix(composer): consume icewind deps from GitHub mirrors to survive Codeberg outages

Backport of #41707 to 10.16.

owncloud CI intermittently fails because Composer must git-clone
icewind/streams and icewind/smb from codeberg.org (no dist archive on
Packagist), and Codeberg regularly returns HTTP 503/504. Every job that
installs Composer dependencies is affected, including "Make local core"
in downstream app builds.

Add VCS repositories pointing at the GitHub mirrors so Composer fetches a
dist zipball via the GitHub API and never touches Codeberg at install time.

Unlike master, 10.16 stays on icewind/smb v3.7.0: the 3.8.x line requires
php >= 8.2 while this branch pins platform php 7.4. Both packages are
re-locked at the exact refs already in the lock (smb e6904cb, streams
cb2bd3e), which the mirrors carry, so this is a pure source repoint with
no version change. No codeberg.org references remain in composer.lock.

Mirrors are kept in sync weekly by DeepDiver1975/codeberg-mirrors.

Signed-off-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>
Co-authored-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>

…0.16] (#41819)

* fix: avoid a deprecation notice when hashing the file cache path on Oracle

Oracle cannot store empty strings, so the file cache converts them to
null before writing a row. For the storage root, whose path is the empty
string, that left md5() being called with null: PHP 8 reports that as a
deprecated implicit null to string conversion, which is noise in the log
and an error under PHPUnit's strict error handling. The stored path_hash
itself was never wrong, because md5(null) coerces to md5('').

Found by running the PHPUnit DB suite against Oracle.

Backport of #41808

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Signed-off-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>

* tmp: debug oracle client on php 7.4

To be dropped before merge.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Signed-off-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>

* fix(composer): consume icewind deps from GitHub mirrors to survive Codeberg outages [10.16] (#41816)

fix(composer): consume icewind deps from GitHub mirrors to survive Codeberg outages

Backport of #41707 to 10.16.

owncloud CI intermittently fails because Composer must git-clone
icewind/streams and icewind/smb from codeberg.org (no dist archive on
Packagist), and Codeberg regularly returns HTTP 503/504. Every job that
installs Composer dependencies is affected, including "Make local core"
in downstream app builds.

Add VCS repositories pointing at the GitHub mirrors so Composer fetches a
dist zipball via the GitHub API and never touches Codeberg at install time.

Unlike master, 10.16 stays on icewind/smb v3.7.0: the 3.8.x line requires
php >= 8.2 while this branch pins platform php 7.4. Both packages are
re-locked at the exact refs already in the lock (smb e6904cb, streams
cb2bd3e), which the mirrors carry, so this is a pure source repoint with
no version change. No codeberg.org references remain in composer.lock.

Mirrors are kept in sync weekly by DeepDiver1975/codeberg-mirrors.

Signed-off-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>
Co-authored-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>

* fix(db): introspect the Oracle schema in a fixed number of queries [10.16]

doctrine/dbal 2.13 describes a schema table by table: for every table it runs
one query for the columns, one for the indexes, one for the foreign keys and
one for the table comment. Every one of those inlines the table name as a
literal, so Oracle cannot share cursors and hard parses each of them at a cost
of a few hundred milliseconds. OC\DB\Migrator asks for the full schema once
per applied migration, which multiplies that per-table cost by the number of
migrations.

Add an OracleSchemaManager that reads the whole data dictionary in a fixed
number of queries and use it for Oracle connections. Reading a 48 table schema
drops from 194 to 6 queries, and occ maintenance:install against Oracle on
PHP 7.4 drops from 43 minutes to 30 seconds. doctrine/dbal itself works this
way from 3.4 onwards, so master is unaffected; its 3.x line cannot be pulled
into 10.16 because it changes public API that third-party apps rely on.

The introspected schema is unchanged. The added test asserts that the batched
result is identical to the stock one according to Doctrine's Comparator, and
that the query count does not grow with the number of tables.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Signed-off-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>

---------

Signed-off-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>
Co-authored-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>

…0.16] (#41819)

* fix: avoid a deprecation notice when hashing the file cache path on Oracle

Oracle cannot store empty strings, so the file cache converts them to
null before writing a row. For the storage root, whose path is the empty
string, that left md5() being called with null: PHP 8 reports that as a
deprecated implicit null to string conversion, which is noise in the log
and an error under PHPUnit's strict error handling. The stored path_hash
itself was never wrong, because md5(null) coerces to md5('').

Found by running the PHPUnit DB suite against Oracle.

Backport of #41808

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Signed-off-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>

* tmp: debug oracle client on php 7.4

To be dropped before merge.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Signed-off-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>

* fix(composer): consume icewind deps from GitHub mirrors to survive Codeberg outages [10.16] (#41816)

fix(composer): consume icewind deps from GitHub mirrors to survive Codeberg outages

Backport of #41707 to 10.16.

owncloud CI intermittently fails because Composer must git-clone
icewind/streams and icewind/smb from codeberg.org (no dist archive on
Packagist), and Codeberg regularly returns HTTP 503/504. Every job that
installs Composer dependencies is affected, including "Make local core"
in downstream app builds.

Add VCS repositories pointing at the GitHub mirrors so Composer fetches a
dist zipball via the GitHub API and never touches Codeberg at install time.

Unlike master, 10.16 stays on icewind/smb v3.7.0: the 3.8.x line requires
php >= 8.2 while this branch pins platform php 7.4. Both packages are
re-locked at the exact refs already in the lock (smb e6904cb, streams
cb2bd3e), which the mirrors carry, so this is a pure source repoint with
no version change. No codeberg.org references remain in composer.lock.

Mirrors are kept in sync weekly by DeepDiver1975/codeberg-mirrors.

Signed-off-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>
Co-authored-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>

* fix(db): introspect the Oracle schema in a fixed number of queries [10.16]

doctrine/dbal 2.13 describes a schema table by table: for every table it runs
one query for the columns, one for the indexes, one for the foreign keys and
one for the table comment. Every one of those inlines the table name as a
literal, so Oracle cannot share cursors and hard parses each of them at a cost
of a few hundred milliseconds. OC\DB\Migrator asks for the full schema once
per applied migration, which multiplies that per-table cost by the number of
migrations.

Add an OracleSchemaManager that reads the whole data dictionary in a fixed
number of queries and use it for Oracle connections. Reading a 48 table schema
drops from 194 to 6 queries, and occ maintenance:install against Oracle on
PHP 7.4 drops from 43 minutes to 30 seconds. doctrine/dbal itself works this
way from 3.4 onwards, so master is unaffected; its 3.x line cannot be pulled
into 10.16 because it changes public API that third-party apps rely on.

The introspected schema is unchanged. The added test asserts that the batched
result is identical to the stock one according to Doctrine's Comparator, and
that the query count does not grow with the number of tables.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Signed-off-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>

---------

Signed-off-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>
Co-authored-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>

…suite against Oracle again [10.16] (#41815)

* ci: run the PHPUnit DB suite against Oracle again

Oracle was never covered by the GitHub Actions workflows: in drone it ran
only in the nightly cron matrix, and that pipeline is not part of the
GitHub Actions setup. Add it as a php-unit.yml matrix entry driven by a
new databases input, served by gvenzl/oracle-free:23-slim-faststart.

Oracle lives in the FREEPDB1 pluggable database, so the installer is
called with a connect string carrying SERVICE_NAME instead of a host.
The oci8 extension is only requested for Oracle jobs, and DB_TYPE is
only exported for Oracle so test-phpunit.sh narrows the run to
--group DB - the other databases keep running the full suite unchanged.

The Oracle job runs on every pull request and shares its display name
with the existing php-unit job, so all databases appear in one group of
checks. It remains a separate job so that a slow or failing Oracle run
cannot cancel the other databases through fail-fast.

Backport of #41808

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Signed-off-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>

* fix(db): introspect the Oracle schema in a fixed number of queries [10.16] (#41819)

* fix: avoid a deprecation notice when hashing the file cache path on Oracle

Oracle cannot store empty strings, so the file cache converts them to
null before writing a row. For the storage root, whose path is the empty
string, that left md5() being called with null: PHP 8 reports that as a
deprecated implicit null to string conversion, which is noise in the log
and an error under PHPUnit's strict error handling. The stored path_hash
itself was never wrong, because md5(null) coerces to md5('').

Found by running the PHPUnit DB suite against Oracle.

Backport of #41808

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Signed-off-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>

* tmp: debug oracle client on php 7.4

To be dropped before merge.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Signed-off-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>

* fix(composer): consume icewind deps from GitHub mirrors to survive Codeberg outages [10.16] (#41816)

fix(composer): consume icewind deps from GitHub mirrors to survive Codeberg outages

Backport of #41707 to 10.16.

owncloud CI intermittently fails because Composer must git-clone
icewind/streams and icewind/smb from codeberg.org (no dist archive on
Packagist), and Codeberg regularly returns HTTP 503/504. Every job that
installs Composer dependencies is affected, including "Make local core"
in downstream app builds.

Add VCS repositories pointing at the GitHub mirrors so Composer fetches a
dist zipball via the GitHub API and never touches Codeberg at install time.

Unlike master, 10.16 stays on icewind/smb v3.7.0: the 3.8.x line requires
php >= 8.2 while this branch pins platform php 7.4. Both packages are
re-locked at the exact refs already in the lock (smb e6904cb, streams
cb2bd3e), which the mirrors carry, so this is a pure source repoint with
no version change. No codeberg.org references remain in composer.lock.

Mirrors are kept in sync weekly by DeepDiver1975/codeberg-mirrors.

Signed-off-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>
Co-authored-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>

* fix(db): introspect the Oracle schema in a fixed number of queries [10.16]

doctrine/dbal 2.13 describes a schema table by table: for every table it runs
one query for the columns, one for the indexes, one for the foreign keys and
one for the table comment. Every one of those inlines the table name as a
literal, so Oracle cannot share cursors and hard parses each of them at a cost
of a few hundred milliseconds. OC\DB\Migrator asks for the full schema once
per applied migration, which multiplies that per-table cost by the number of
migrations.

Add an OracleSchemaManager that reads the whole data dictionary in a fixed
number of queries and use it for Oracle connections. Reading a 48 table schema
drops from 194 to 6 queries, and occ maintenance:install against Oracle on
PHP 7.4 drops from 43 minutes to 30 seconds. doctrine/dbal itself works this
way from 3.4 onwards, so master is unaffected; its 3.x line cannot be pulled
into 10.16 because it changes public API that third-party apps rely on.

The introspected schema is unchanged. The added test asserts that the batched
result is identical to the stock one according to Doctrine's Comparator, and
that the query count does not grow with the number of tables.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Signed-off-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>

---------

Signed-off-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>
Co-authored-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>

* ci: drop the temporary Oracle debug instrumentation

The Oracle job carried two pieces of throwaway diagnostics that were only
there to find out why maintenance:install took an hour on this branch: a
"TEMPORARY DEBUG - Oracle client" step probing oci8, the instant client
and a raw connect, and a v$session/v$sql sampler that wrapped the install
in a background job with a 90 minute kill timeout and printed the top SQL
afterwards.

That question is answered - it was doctrine/dbal 2.13's per table schema
introspection, fixed by #41819 - and the install is back to under a
minute, so the instrumentation has no reason to stay. Installing on
Oracle is a plain "php occ ${install_cmd}" again, like every other
database.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Signed-off-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>

---------

Signed-off-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>
Co-authored-by: Thomas Müller <323649642+oc-tmueller@users.noreply.github.com>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>