Pending translations lost during repository sync due to flawed needs_commit_upstream() logic

[dxdx123]

Describe the issue

Weblate overwrites uncommitted local translation changes when syncing from Git repository, causing translator work to be permanently lost. This happens because the needs_commit_upstream() method only checks for Git file changes but ignores pending translation changes in the Weblate database.

I already tried

• I've read and searched the documentation.
• I've searched for similar filed issues in this repository.

Steps to reproduce the behavior

1. User makes a translation change in Weblate (e.g., changes a term from "RuffLace" to "ExcRuffle")
2. The change is saved in Weblate's database but not yet committed to Git (still marked as pending=True)
3. Repository sync is triggered (either automatically by hourly task or manually)
4. needs_commit_upstream() returns False because no Git files have changed
5. Weblate skips the commit step and directly syncs from Git
6. Translation files are read from Git, overwriting database changes
7. User's translation work is permanently lost

Expected behavior

Expected behavior

Weblate should protect uncommitted local changes during repository sync by:

1. Checking for pending translations in the database before syncing
2. Automatically committing pending changes before performing repository updates
3. Never overwriting translator work that hasn't been pushed to Git

Actual behavior

Pending translation changes are silently overwritten and lost during repository sync, with no warning or recovery mechanism.

Root cause analysis

The issue is in the needs_commit_upstream() method in weblate/trans/models/component.py:

def needs_commit_upstream(self) -> bool:
    """Detect whether commit is needed for upstream changes."""
    changed = self.repository.get_changed_files()  # Only checks Git repository
    if self.uses_changed_files(changed):
        return True
    # ... but never checks for pending translation changes in database!

This method only examines Git repository file changes, completely ignoring pending translation units in the Weblate database that haven't been written to files yet.

Evidence from our logs

We observed this behavior in our production instance:

• Last normal commit: 21 hours ago (18:29:24)
• Translation lost: User "anglezhang" modified term 20 hours ago
• Repository sync: 18 hours ago (8:30 PM) - overwrote the pending change
• Next commit: 4 hours ago (11:24:08) - 17-hour gap indicates sync disruption

Suggested fix

Modify needs_commit_upstream() to also check for pending database changes:

def needs_commit_upstream(self) -> bool:
    """Detect whether commit is needed for upstream changes."""
    changed = self.repository.get_changed_files()
    if self.uses_changed_files(changed):
        return True
    
    # Check for pending translation changes in database
    if self.pending_units.exists():
        return True
        
    for component in self.linked_childs:
        if component.uses_changed_files(changed) or component.pending_units.exists():
            return True
    return False

Impact

This is a critical data loss bug that can result in:

• Loss of translator work and productivity
• Decreased trust in the platform
• Potential compliance issues for professional translation workflows
• Silent failures without user notification

Environment

• Weblate version: Multiple versions affected (we've seen this in production)
• The issue exists in the current codebase architecture
• Affects all Weblate installations using automatic repository sync

Additional context

This issue particularly affects workflows where:

• commit_pending_age is set to longer periods (default 24 hours)
• Automatic repository sync is enabled (AUTO_UPDATE)
• Multiple translators work on the same component
• Integration with external Git repositories is used

The problem is exacerbated by Weblate's hourly sync schedule (update_remotes task), which can trigger before the 24-hour commit timeout expires.

Screenshots

Pending translation changes are silently overwritten and lost during repository sync, with no warning or recovery mechanism.

Exception traceback

How do you run Weblate?

Docker container

Weblate versions

weblate 5.10.4

Weblate deploy checks

Additional context

No response

[nijel]

The purpose of needs_commit_upstream is to figure out whether there are any changes to the translation files in the upstream repository. In case the changes do not touch translation files, it should be safe to merge it, as no translation changes are parsed in this case (and thus cannot be lost).

So what you most likely observe is that needs_commit_upstream doesn't detect upstream updates as changing the translation files.

Can you please elaborate a bit more on your setup? What file format do you use? What component configuration? Can you share logs from the overwrite of the translations?

PS: The update_remotes is scheduled hourly but triggers once per 24 hours for each component with a remote repository.

[dxdx123]

I can't find this commit in the upstream Git repository, which means it hasn't been pushed there yet. What's even more concerning is that I can't find any record of it, even with a time difference of over ten days.

[nijel]

The history (Translation changed followed by String updated in the repository) indicates that either Weblate has issues writing out changes to the translation file or detecting the upstream changes. What actually happened underneath cannot be fully seen from this.

I will repeat my question: Can you please elaborate a bit more on your setup? What file format do you use? What component configuration? Can you share server logs from the overwrite of the translations?

[dxdx123]

Detailed Configuration Information

File Format and Component Configuration

File Format: XLIFF (.xliff)
Component Type: Glossary
Component Name: project_glossary_approved_sourceCN
File Path: components/project_glossary_approved/sourceCN/en.xliff (and other language files)

Core Issue Description

We encountered two related data loss issues:

Issue 1: User Translation Overwritten by Anonymous

• Time: 2025-07-21
• Phenomenon: User "user1"'s modifications were cleared by "Anonymous" user
• Specific Behavior: Term changed from Friendshine RuffLace Sleeve Covers → Friendshine ExcRuffle Sleeve Covers → reverted to Friendshine RuffLace Sleeve Covers

Issue 2: GitLab Webhook Triggered Data Overwrite

• Trigger Mechanism: GitLab webhook automatic sync
• User Identity: Displayed as "Anonymous" (actually webhook-triggered background task)

Key Log Evidence

1. GitLab Webhook Trigger Log (2025-07-21 10:09:12)

[2025-07-21 10:09:12,805: INFO/1715935] received GitLab notification on repository company/l10n/weblate_repo/weblate_project, URL https://git.company.com/company/l10n/weblate_repo/weblate_project, branch production, 1 matching components, 1 to process, 4 linked

[2025-07-21 10:09:12,805: INFO/1715935] GitLab notification will update Project XYZ/project_uiprefab_sourceCN

# HTTP request log
nginx stdout | 100.127.128.72 - - [21/Jul/2025:02:09:12 +0000] "POST /hooks/gitlab/ HTTP/1.1" 200 234 "-" "GitLab/18.2.0-ee"

# Async task triggered
celery-celery stderr | [2025-07-21 10:09:12,987: INFO/MainProcess] Task weblate.trans.tasks.perform_update[a1b2c3d4] received

2. Normal Commits Before Batch Push (10:08:45-10:08:59)

[2025-07-21 10:08:45,436: INFO/1715935] project/project_uiprefab_sourcecn: scheduling push
[2025-07-21 10:08:45,436: INFO/1715935] project/project_glossary_approved_sourcecn/th: committed ['/app/data/vcs/project/project_uiprefab_sourcecn/components/project_glossary_approved/sourceCN/th.xliff'] as User2 <user2@company.com>

[2025-07-21 10:08:47,411: INFO/1715935] project/project_uiprefab_sourcecn: scheduling push  
[2025-07-21 10:08:47,411: INFO/1715935] project/project_glossary_approved_sourcecn/zh_Hant: committed ['/app/data/vcs/project/project_uiprefab_sourcecn/components/project_glossary_approved/sourceCN/zh-Hant.xliff'] as User2 <user2@company.com>

Root Cause Analysis

Our analysis shows the issue occurs in the following flow:

1. Batch Push Trigger: Weblate batch pushes multiple pending translations to GitLab (10:08:45-10:08:59)
2. Webhook Trigger: GitLab detects the push and sends webhook to Weblate (10:09:12)
3. Sync Execution: Webhook-triggered async task executes do_update() without request context
4. User Identity Loss: Due to missing request, user is set to None, displayed as "Anonymous"
5. Data Overwrite: Sync process overwrites database changes with Git file content

Key Code Paths

Based on our code analysis, the issue occurs in:

# weblate/trans/views/hooks.py:211
perform_update.delay("Component", obj.pk)  # No user context passed

# weblate/trans/tasks.py:52-74  
def perform_update(cls, pk, auto=False, obj=None):
    obj.do_update()  # request=None

# weblate/trans/models/translation.py:373
def check_sync(self, request=None, ...):
    user = None if request is None else request.user  # None leads to Anonymous

Timeline of Events

• 17 hours ago: Normal commits stopped
• GitLab Push: Batch push multiple translations to GitLab
• Webhook Sync: Immediately triggers sync, overwrites data
• 4 hours ago: Normal commits resumed (17-hour gap anomaly)

Environment Information

Weblate Version: 5.10.4
Deployment: Docker container
Git Integration: GitLab webhook automatic sync
File Format: XLIFF
Component Type: Glossary

The core of this issue is that webhook-triggered sync lacks proper protection mechanism for pending changes, resulting in user's uncommitted translations being overwritten.

Reproduction Conditions

This issue inevitably occurs under the following conditions:

1. User makes translations in Weblate but hasn't committed to Git yet
2. Other translations are batch-pushed to GitLab repository
3. GitLab webhook triggers Weblate sync
4. Sync process overwrites pending changes in database

[nijel]

Did the commits pushed to GitLab contain problematic changes? To me it seems that here is the problem. Maybe you're hitting #2482?

PS: Does it happen on a current Weblate version as well?