Do for the CSV input module what commit 08f8421a9e did for VCD. Check
the channel list for consistency across re-imports of the same file.
This addresses the CSV part of bug #1241.
The cleanup() routine gets invoked upon shutdown, as well as before
re-importing another file. The cleanup() routine must not release
resources which get allocated in the init() routine, as the init()
routine won't run again in the module's lifetime. The cleanup() routine
must void those context fields which get evaluated in the next receive()
calls.
Keep (part of) previous results around when the VCD input module gets
reset, and check the header of the next import against previous runs to
avoid issues in applications. This addresses the VCD specific part of
bug #1241, and resolves the remaining part of bug #1306.
This addresses part of bug #1306. The reset() method of the VCD input
module was incomplete, and did not process new data upon second read.
Improve robustness and add missing reset instructions. Void invalid
pointers and avoid NULL dereferences in cleanup paths.
Applications might pass NULL for the buffer, and input modules might
accept it (or just cope). Eliminate a potential NULL dereference in
the emission of diagnostics messages.
With gcc 8 this yielded:
src/input/wav.c: In function ‘receive’:
src/input/wav.c:345:51: warning: ‘%d’ directive output may be truncated writing between 1 and 10 bytes into a region of size 6 [-Wformat-truncation=]
snprintf(channelname, sizeof(channelname), "CH%d", i + 1);
^~
src/input/wav.c:345:48: note: directive argument in the range [1, 2147483647]
snprintf(channelname, sizeof(channelname), "CH%d", i + 1);
^~~~~~
src/input/wav.c:345:5: note: ‘snprintf’ output between 4 and 13 bytes into a destination of size 8
snprintf(channelname, sizeof(channelname), "CH%d", i + 1);
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The VCD specification requests that timestamps will strictly increase as
one advances through the file. Add another check where the previous
implementation resulted in a tight loop and made the application stall.
Do print an error message and abort file processing in that case.
This fixes bug #1250.
Split the creation of channels and groups as well as the creation of the
session feed buffer into separate routines. Re-allocate the buffer after
reset but do not re-create the channels and groups.
This implementation assumes that after reset() of the input module, the
very same set of channels (including their order, names and enabled
state, as well as group membership) results from processing the
subsequently fed file content. Reading rather different configurations
from the same input file by means of repeated reset and re-read may not
work as expected.
Explicitly check GString pointers for validity before calling glib
routines that would access string content. This silences an assertion
error for a non-fatal runtime condition:
(process:17044): GLib-CRITICAL **: g_string_free: assertion 'string != NULL' failed
Rephrase common input reset logic such that additional common code can
execute after the individual module's reset callback. No behaviour has
changed, the module's reset callback still is optional, and identical
log output gets emitted.
Do clear the sdi_ready flag in the common sr_input_reset() routine, so
that all input modules will parse header information again before
processing sample data when subsequent calls receive new file content.
Void the input module's receive() buffer from common reset code. This
unbreaks the feature of re-reading previously consumed input files.
Extend comments in the common reset and free code paths, which involve
the modules' reset and cleanup routines, which interact in non-trivial
ways. Discuss the responsibilities of common and individual routines, to
remain aware during maintenance.
The chronovu_la8 input module is capable of handling either file type,
generated by LA8 or LA16 vendor software. Automatic detection is not
available, but user provided channel counts work fine.
Adjust the input module's name and description, and claim support for
.kdt as well as .kd1 files.
Add developer notes on implementation details, the file content's
constraints, and potential future adjustment.
The file format is funny, a data part is leading (fixed size) and a
"header" part follows. The previous implementation sent the header part
to the session, too, pretending it was part of the data. This change
limits the number of samples that get sent to the session.
Comment on the file layout and header fields while we are here. This
information got lost in commit 02e24c0ce0 when the input module got
converted from do-it-yourself file operations to having intrinsic
handler routines invoked from common logic which handles the file.
Prefer the more portable UINT64_C() macro over the UL suffix when a
literal needs to span "more than 31 bit positions". Adjust other
locations for consistency across the source file.
Commit 8c4bff1d25 introduced a routine which prints what mostly should
be text, and avoids non-printable characters. This implementation used an
incorrect format string, which could result in data getting written past
the end of a buffer. Fix the format string.
Reported by scan-build:
src/input/logicport.c:1186:18: warning: initialization from incompatible pointer type [-Wincompatible-pointer-types]
.format_match = format_match,
^~~~~~~~~~~~
Log determined header format when parsing trace32 file headers. Keep
another log message during regular processing, but silence it during
format match.
Use more appropriate data types for local and instance variables. Adjust
format strings accordingly.
There were several issues with the previous implementation of the logic
which parses trace32 input files: A string comparison was inverted, and
compared a seven character literal to a five character copy of the input.
One more character was trimmed before CTRL-Z (the CP/M EOF), which often
is SPACE in input files, but might be excessive on other input files.
Replace a DIY character search while we are here. Use symbolic names for
special characters. Factor out a test and memory release, to simplify
error code paths in the remaining logic. Extend comments.
There is not much point in log messages about format mismatch in the
auto detect phase, where each module gets queried in turn and most are
supposed to not match.
Do not print non-printable characters in log messages. Those could occur
in the format detection phase, or in the regular processing of input
files that either are unexpected or invalid (or is there a binary header
format variant even?).
Although the file format handled by this input module appears to be of
fixed size (8MiB plus 5 more bytes), it's more reliable to use a data
type for the file size that is larger than "an int". Although off_t
would be most portable, use uint64_t to match the code which passes the
parameter to the input module.
The previous implementation already was inconsistent (used to allocate
255 bytes and claimed that 128 bytes were sufficient). While existing
formats already required more than a few bytes of input (regular VCD
files' header sections exceed 255 bytes length).
Increase the buffer size that gets passed to input modules' match
method. Use 4MiB for consistency with other locations. Do not enforce a
minimum size, as there are valid input files which are shorter than 128
bytes. Auto-detection failed on those.
This addresses part of bug #1200.
When users don't specify the input format, applications can try to have
the format auto-detected. Some of the tests were weak and could result
in false positives.
Add a 'confidence' parameter to the input modules' format_match()
method. Claim high confidence (1) for those formats which can check
magic strings or the presence of essential keywords (vcd, wav). Claim
medium confidence (10) for those formats which happened to mis-detect
inputs in the past (trace32_ad). Claim weak confidence (100) for those
formats which totally lack reliable conditions and rather guess than
detect (chronovu_la8).
Prefer the best match in public scan routines. Return at most one module
to callers even if multiple modules matched, to keep the current API.
This addresses part of bug #1200.
Implement an input module for .lpf files, the "LogicPort File" format of
the Intronix LA1034 vendor software. This version supports wires with
enabled and inverted state, compressed samples, signal names, signal
groups (but not multiple assignment), and automatic format detection.
The logic which determines whether "the file header" was completely
received, and sample data can get sent to the session, implements an
assumption in the absence of a better and more reliable condition.
Invalid input gets rejected, but diagnostics is rather limited. Since
all channels get to be the member of a channel group, either specified
by the user in the input file, or arranged for in the input module, the
"missing separator" part of bug 1186 has become obsolete.
When auto-detection tries to find the input module that can handle a
file, unexpected input format should be considered non-fatal. Only emit
error messages when process_header() got invoked from process_buffer()
after initialization. Emit debug messages in early stages where no input
module context exists yet.
When processing of large VCD input files was spread across multiple
parse_contents() invocations, the resulting sigrok stream of sample data
had gaps in them and total timing was off. For instance 74ms of input
data were interpreted as spanning some 600ms or 300ms, depending on the
number of channels in the input stream.
Move the "previous timestamp" variable to the input module context. This
eliminates the inappropriate gaps and fixes the translation of VCD file
timestamps to sigrok sample numbers.
This fixes bug #1075.
../src/input/wav.c: In function ‘send_chunk’:
../src/input/wav.c:200:2: warning: ‘memset’ used with length equal to number of elements without multiplication by element size [-Wmemset-elt-size]
memset(fdata, 0, CHUNK_SIZE);
^~~~~~
../src/input/raw_analog.c: In function ‘init’:
../src/input/raw_analog.c:133:31: warning: ‘%d’ directive output may be truncated writing between 1 and 10 bytes into a region of size 6 [-Wformat-truncation=]
snprintf(channelname, 8, "CH%d", i + 1);
^~
../src/input/raw_analog.c:133:28: note: directive argument in the range [1, 2147483647]
snprintf(channelname, 8, "CH%d", i + 1);
^~~~~~
../src/input/raw_analog.c:133:3: note: ‘snprintf’ output between 4 and 13 bytes into a destination of size 8
snprintf(channelname, 8, "CH%d", i + 1);
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
When the processing of columns of text lines detected errors, the loop
was aborted and the routine was left, but allocated resources were not
freed. Fix the remaining memory leaks in the error code paths.
The constant at the top of the source file is the number of samples in a
datafeed submission chunk. The previous implementation erroneously made
it the size in bytes. There is no need to round down the buffer size
according to the unit size.
The previous implementation sent one sigrok session datafeed packet per
processed CSV line. This is rather inefficient for the CSV input module,
and triggers a dramatic performance loss in the srzip output format.
Communicate up to 128K samples within one datafeed packet. This fixes
bug #695.
Factor out repeated calculation of the unit size which is derived from
the channel count. Fix a minor memory leak in an error path while we are
here. (Other memory leaks in rare error paths remain with this commit.)
Suggested-By: Elias Oenal <sigrok@eliasoenal.com>
On the Windows platform it appears to be popular to _not_ terminate the
very last line in a text file. Which results in an unmet constraint in
the CSV input module and an internal exception in PulseView which aborts
program execution.
Cope with the absence of the text line termination sequence at the very
end of the input stream. Keep all other checks in place, such that only
completely received text lines get processed.
This fixes bug #635.
"Document" the current state of the implementation in the CSV input
module's source code. Discuss how text handling is non-trivial, which
approaches are available and how they have drawbacks.
Mention the lack of support for the import of analog data as well.
The CSV input module supports variable length end-of-line encodings
(either CRLF, or CR, or LF). When a bunch of accumulated text lines got
processed, do skip the corresponding number of characters after the end
of the last processed line.
This fixes one of the issues discussed in bug #635.
The input module runs receive() and end() invocations which end up
calling process_buffer(). It's perfectly legal to call the process
routine with an empty accumulation buffer, especially when the process
routine was called from end().
This fixes a condition where PulseView raised a fatal error at the end
of a completed successful import.
Reported-By: Sergey Alirzaev <zl29ah@gmail.com>
Move an independent test for single/multi column operation out of a code
path that checked for and then processed text lines. This commit does
not change behaviour, but prepares a subsequent commit.
Factor out a magic string literal which held a delimiter set yet could
be mistaken for an (assumed) fixed termination string. Concentrate the
determination of the end-of-line text encoding as well as the resulting
set of possible deliminters in one nearby location. The symbolic name
for the delimiter set eliminates the doubt on its purpose.
The default so far was 0, which meant there would be no significant
digits at all, yielding results that looked strange/wrong to the user.
Long-term all remaining drivers should be fixed to use the actual,
correct digits and spec_digits values according to the device's
capabilities and/or datasheet/manual. Until that is done, a default
of digits=2 is used as a temporary workaround.
This fixes the remaining parts of bug #815.
Some of the standard helper functions take a log prefix parameter that is
used when printing messages. This log prefix is almost always identical to
the name field in the driver's sr_dev_driver struct. The only exception are
drivers which register multiple sr_dev_driver structs.
Instead of passing the log prefix as a parameter simply use the driver's
name. This simplifies the API, gives consistent behaviour between different
drivers and also makes it easier to identify where the message originates
when a driver registers sr_dev_driver structs.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
This makes the code shorter, simpler and more consistent, and also
ensures that the (same) debug messages are always emitted and the
packet.payload field is consistently set to NULL always, etc.
According to the infos I have, VCD files should be plain ASCII, but we
got report of a version adding a UTF8 BOM at the beginning of the file,
so we need to skip it.
This fixes bug #755.
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
Accept (and set to default) 0 for numchannels which means 'unlimited'.
I think it is convenient to read all channels of a vcd file by default.
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
Benefits:
* only channels really used in vcd will be added
* we can give them the proper name found in the vcd file
* less code :)
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
Current code moves the identifier string one byte to the front to
overwrite the bit value, so 'tokens[i]' is a string to compare against
the desired value. This copying is unnecessary, just pass a properly
setup pointer.
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
No need to bail out on vectors. As long as there are tokens left, we can
try to parse the rest. Also, print a message so the user knows what's
going on. Here is a testcase vcd:
$timescale 1 ns $end
$var wire 1 n0 addr_0 $end
$var wire 1 n1 addr_1 $end
$enddefinitions $end
#0
0n0
b1 n1
#1
1n0
b0 n1
#2
0n0
b1 n1
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
If we hit the missing identifier case, then we reached the end of the
token list. So, we should break out of the loop, and not continue.
Otherwise we will go past the end of the array as this minimal testcase
shows:
$timescale 1 ns $end
$var wire 1 n0 addr_0 $end
$enddefinitions $end
1
gives:
$ ./sigrok-cli -I vcd -i no_mod.vcd -O vcd -o /tmp/o.vcd
Segmentation fault
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
A $var block can have an optional index item which looks like '[<sth>]'.
Parse it, too, and append it to the channel name.
This fixes bug #322. A first version was posted by Simon Richter. This
version is rebased and simplified.
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
As we're downsampling, several record time stamps can match the specified
trigger time. For this reason, it's possible that several trigger packets
are sent when a file is loaded. This prevents the issue and sends a
trigger packet only on the first matching record.
The sr_input_dev_inst_get API documentation guarantees an input is fully
initialized as soon as the device instance is returned. An sdi
implementation should not set sdi_ready any earlier.
This fixes parts of bug #387.
Gerhard Sittig
Soeren Apel
Martin Ling
Uwe Hermann
Lars-Peter Clausen
Wolfram Sang
Stefan Brüns
Petteri Aimonen
Bert Vermeulen