summaryrefslogtreecommitdiff
path: root/src/settings.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/settings.cpp')
-rw-r--r--src/settings.cpp29
1 files changed, 28 insertions, 1 deletions
diff --git a/src/settings.cpp b/src/settings.cpp
index 6445bfbe..f5ad83a7 100644
--- a/src/settings.cpp
+++ b/src/settings.cpp
@@ -470,7 +470,34 @@ const DiagnosticsSettings& Settings::diagnostics() const
QSettings::Status Settings::sync() const
{
m_Settings.sync();
- return m_Settings.status();
+
+ const auto s = m_Settings.status();
+
+ // there's a bug in Qt at least until 5.15.0 where a utf-8 bom in the ini is
+ // handled correctly but still sets FormatError
+ //
+ // see qsettings.cpp, in QConfFileSettingsPrivate::readIniFile(), there's a
+ // specific check for utf-8, which adjusts `dataPos` so it's skipped, but
+ // the FLUSH_CURRENT_SECTION() macro uses `currentSectionStart`, and that one
+ // isn't adjusted when changing `dataPos` on the first line and so stays 0
+ //
+ // this puts the bom in `unparsedIniSections` and eventually sets FormatError
+ // somewhere
+ //
+ //
+ // the other problem is that the status is never reset, not even when calling
+ // sync(), so the FormatError that's returned here is actually from reading
+ // the ini, not writing it
+ //
+ //
+ // since it's impossible to get a FormatError on write, it's considered to
+ // be a NoError here
+
+ if (s == QSettings::FormatError) {
+ return QSettings::NoError;
+ } else {
+ return s;
+ }
}
QSettings::Status Settings::iniStatus() const