| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
TDevInfResultsCommand::shrinkCommand()
|
|
TDevInfResultsCommand::shrinkCommand()
|
|
In (probably mostly unused code?) the previous boolean result
of getIncomingState() was compared against the enum, which looks
wrong. Probably getIncomingState() was meant to return the enum.
gcc6 warned about this:
synccommand.cpp:4277:36: error: comparison of constant '1' with boolean
expression is always false [-Werror=bool-compare]
fSessionP->getIncomingState()>psta_init, // alerted only if init complete
|
|
Recent gcc and/or C++14 changed the definition of abs() such that
applying it to a 16bit integer resulted in a float, leading to:
itemfield.cpp:1491:27: error: invalid operands of types
'__gnu_cxx::__enable_if<true, double>::__type {aka double}' and 'const int' to
binary 'operator%'
(uInt16)(abs(moffs) % MinsPerHour)
|
|
This fixes:
# warning "_BSD_SOURCE and _SVID_SOURCE are deprecated, use _DEFAULT_SOURCE"
|
|
libical 2.0 has an API call to enable the portable timezone definitions
required by libsynthesis (one entry for summer, one for winter). When
possible (= compiling without compatibility hack, libical recent enough)
we should use that.
|
|
The source code was intentionally formatted by the upstream
authors so that statements sometimes align with surrounding
code even if indention according to program flow would be
different.
Instead of fixing large amounts of code, better just ignore
this warning.
|
|
Despite what the comment for the method said, the parameter was dereferenced
unconditionally and thus was not allowed to be NULL. Fixing the comment and
removing the redundant NULL check. Found with clang.
|
|
va_start() may allocate resources and thus has to be matched by
va_end(). Found with cppcheck 1.67.
|
|
va_start() may allocate resources and thus has to be matched by
va_end(). Found with cppcheck 1.67.
|
|
clang 3.5.0 warns about checking the "this" pointer because it cannot
be NULL. This safe-guard against incorrect usage of the class indeed
seems to be impossible, so remove the check.
|
|
Implicit conversion of long to int when calling abs() triggers a
compiler warning in clang 3.5.0.
|
|
The / in TZID values like Europe/Paris caused the code from "MIME
parser + encoder: no backslash quoting in parameter values" (8a7356c,
libsynthesis_3.4.0.47+syncevolution-1-4-99-2~6) to quote the
TZID value.
The Funambol OneMedia iCalendar 2.0 parser fails to handle this;
instead it imports events without applying the time zone.
It is not necessary to use quotation marks around values containing a
slash, so add the slash to the white list of characters which do no
trigger quoting.
|
|
|
|
This stricter version of RELAXEDCOMPARE() reports a difference also
when only the order of entries differs. This is easier to implement,
faster at runtime (min(n,m) comparisons instead of (n*m)), and takes
into account that the order may be relevant.
|
|
Caching mode in SyncEvolution only worked if the incoming item matched
the stored one exactly. Otherwise changing the order of items
(e.g. TEL followed by EMAIL instead of EMAIL followed by TEL) caused
arrays to be different (because of the shared LABEL array), and that
difference in the arrays led to rewriting items even if nothing
changed.
Because of the decode/encode step before storing a new item, caching
only worked reliably when the incoming item was also decoded and
encoded once, which is unnecessary overhead and only worked when using
a storage with the same encoding as the local sync.
A better solution is to use a more intelligent merge script:
- check whether arrays are identical except for ordering (done
with the new RELAXEDCOMPARE())
- ignore arrays which were found to be similar enough (by
passing a new parameter listing these fields to MERGEFIELDS().
In essence, MERGEFIELDS() then is only asked to check the remaining
fields or merge arrays if relevant differences were found.
At the moment, RELAXEDCOMPARE() only checks that entries in the arrays
match. It does not check that their order is the same. This could
(should!) be added and would also make the comparison faster when
there are differences (because it only needs to compare the next
non-empty entries).
|
|
and bug fixes
|
|
In one particular case, a pending command that needs more data, the engine
crashed because delayed command execution code path deleted the command still
needed for the next chunk (found in Client::Sync::eds_contact_eds_memo::testLargeObject).
|
|
SyncEvolution will intercept alloc/free during local sync to put
message buffers into shared memory and thus needs this feature
enabled.
|
|
This additional indirection allows users of the library to replace the
default implementation from libc.
The feature is optional. In libsynthesis it can be enabled in
target_options.h. Users of libsynthesis do not have access to
libmem.h, so they need to provide their own declaration of the
function pointers.
|
|
A signed long for the memory functions is inconsistent with
systems that have a standard malloc/realloc. Better use size_t.
|
|
Some code only compiles cleanly when MemSize_t is signed. Otherwise there are
signed/unsigned comparison and printf format warnings. Avoid those by casting
the values which are known to be positive to MemSize_t or long in debug
output.
|
|
Some functions were defined in both sml.h and libmem.h. Let's
use libmem.h as the authoritative place and eliminate the copies
in sml.h. Will be necessary when enhancing libmem.h.
|
|
The goal is to leave binfile completely untouched, including its "last
modified" time stamp, if nothing changes.
There are cases where the upper layers will call
TBinFileBase::updateRecord() and write the same data that already
is in the file. Instead of catching those case-by-case, let's check
at the lowest level.
On btrfs (and possibly other file systems), ftruncate() bumps the
"last modified" time stamp of a file even if the requested size
matches the existing one. Avoid that by skipping the ftruncate() when
the file already has the desired size.
|
|
During a normal sync where nothing changed, only the header gets
updated. This change is not critical and thus does not have to be
flushed to disk unless also some entries get added or updated.
The advantage is that when SyncEvolution detects a sync where nothing
changed on either side and skips the client's session shutdown, the
.bfi is left unchanged, which reduces flash wearout.
To detect item changes, a brute-force byte comparison is used. This
requires less changes to the code and is less error-prone than adding
"modified=true" to all places where "existingentries" gets modified.
|
|
When the configure auth type needs no nonce, don't generate one
and return NULL immediately, like newChallenge() would do.
That avoids unnecessary disk writes for storing the new nonce, which
is important for local syncs in IVI. It also avoids debug output about
a challenge that would not get sent.
|
|
Package name and version should be passed to AC_INIT according to
current Automake documentation.
In addition, we need to pass subdir-objects as option to
AM_INIT_AUTOMAKE to avoid warnings about the location of our object
files (first seen with automake 1.14.1).
|
|
|
|
When the client already knows that it is never going to sync again or
the next sync will have to be a slow sync anyway, then the binfile
client files are not needed. Suppressing writing of them is useful on
devices with flash memory to avoid flash wearout. The specific use
case is PBAP caching in a car's head unit.
When setting "/dev/null" as datadir, the platform layer will avoid
creating the files and instead keep their content in memory. This is
required (and sufficient) for syncing once, which will close and later
read from some files at least once. Therefore the data cannot be
thrown away completely.
|
|
clang 3.4 warned about unused global variables.
|
|
Item operations returning LOCERR_AGAIN do not need to be logged
as error. This will trigger logging elsewhere, so don't log it at
all in the binfile layer.
|
|
clang 3.4 points out that the ! in !aQuoteMode!=qm_none in
quoteStringAppend() only gets applied to aQuoteMode. Explicit
brackets silence the compiler warning without changing the
expression.
The compiler indeed points out some unusual code. The if branch is
taken for aQuoteMode == qm_none (probably what was intended) because
!qm_none != qm_none with qm_none == 0 and not taken for all other
values where !aQuoteMode is false and thus equal to qm_none.
It probably would be simpler to write "if (aQuoteMode == qm_none)" but
I don't want to change the meaning of the code.
|
|
String fields also get used for arbitrary binary data, like
photos. In that case we need to compare the entire std::string,
not just the part before any embedded null byte.
This gets done using std::string::compare. Case-insensitive
comparison still uses C strucmp() and thus should never be used
for non-string data.
|
|
gcc 4.9 found a flaw in the code path for updating an item when the DB
detects a duplicate: some variables where not preserved across
function restarts.
|
|
gcc 4.9 warns about initializing a boolean with NULL. Probably
should have been false to start with.
|
|
fIsStringBLOB is only used in pluginapids. Found by gcc 4.9.
|
|
gcc 4.9 warns that "if (ext == 255)" is always false due to casting
rules. Need to convert 255 into the same type as ext, as done
elsewhere.
|
|
gcc 4.9 warns about the unused local variables. sizeof() does not need
real variables, so use a casted NULL pointer instead and get rid of
the variables and the warning.
|
|
gcc 4.9 warns about uninitialized tctx. It doesn't seem to notice that
the code using it is not reached unless ISO8601StrToTimestamp()
succeeds and initializes the variable first.
Suppress the warning by initializing to zero first.
|
|
cppcheck 1.65 warns about uninitialized variables in the "retry function" code
blocks. These are false positives because the code is only reached after
initializing them and/or restoring unitialized variables would not matter.
Have to add inline suppressions to get rid of the warnings.
|
|
cppcheck 1.65 warned about uninitialized variables because it did not fully
understand the indirect "if no file, then error, if no error, then use file"
logic.
Simplifying the code to check the file directly avoids the false positives
and also is easier to read for humans.
|
|
sharedfield
|
|
When two properties (TEL and X-ABLabel in this example) share the same group
field, then storing the property where a group tag was used for the first time
must not reuse unassigned group field entries.
There was some logic for that already (see "someGroups"), but it failed to
handle this case. This approach here uses brute-force instead of trying to me
smart: the reuse check in the n<maxrep for loop will fail repeatedly at the
beginning until reaching the end of the group field array.
|
|
In an app which starts a session with some already existing logger, the
session logger doesn't have enough information (no output channel, no log
path) to create thread-specific loggers. It must let the parent logger
do that.
|
|
The variable assignment became redundant when removing the backslash escaping
support for parameters in "MIME parser + encoder: no backslash quoting in
parameter values".
|
|
The temporary variable was never freed.
|
|
The URLDECODE() function unescapes all %XX sequences. URLENCODE()
escapes everything which is not alphanumeric.
|
|
RFC 2425 (MIME DIR) and RFC 6350 (vCard) do not describe backslash
escaping for special characters in parameter values. Only the more
limited quoted-string (double quotes at start and end, no line breaks,
no double quotes inside value) is specified.
The following two examples both contain a literal backslash:
URL;X-ABLabel="Custom-label6 Backslash \":http://custom.com
X-ABRELATEDNAMES;X-ABLabel="custom-label5 Umlaut ä Semicolon ; Backslash \ end of label":custom relationship
This commit limits backslash escaping to parsing and generating
property values. Backslashs in parameters are stored literally during
parsing. A quoted string parameter value is used for every value that
is more complex than alphanumeric plus underscore and hyphen.
In particular spaces (while allowed in unquoted values) are only
generated as part of quoted strings because the Evolution Data Server
parser had problems with them (fixed in EDS 3.10).
|
|
Traditionally, the value of a property and its parameter values get
stored in fields reserved for that property such that there is no
overlap with other properties. In such a scenario it is enough to
check for unused fields used for the property value, because the
corresponding parameter fields are guaranteed to be available once a
repetition is found.
However, once the same field is used for different properties the
check and creation becomes more complicated: when parsing properties,
the shared field must also be checked. It is not enough for it to be
empty, it also must not be in use at all, because otherwise a value might
get added to the repetition used already by some other property.
This check only makes sense (and only works) for array fields. It is
disabled by default and needs to be enabled explicitly with the new
parameter attribute "sharedfield".
Storing values must ensure that the shared array field gets resized if
the parameter was not set, otherwise the check would fail to detect
used repetitions.
The effect of using a shared field is that arrays contain a lot of
unassigned values because index overlaps between otherwise unrelated
properties get avoided.
This feature was needed for converting between a profile with custom
labels in a X-ABLabel property attached to various contact properties
(ADR, TEL, X-ABDate, etc.) via group tags and a profile using a
property parameter instead. The label array field is the shared field
in this case.
|
|
When using fork+exec on Linux, open files get inherited by the child
process unless explicitly marked as FD_CLOEXEC. We should set that
flag at least for files which remain open (like the log file when
dbgflush_openclose is not active). A truly thread-safe approach would
have to use open(O_CLOEXEC), but setting the flag later is sufficient
for us.
This change is limited to Linux to avoid interfering with other
platforms.
|
