0.9.0 #81

adfVolReadBlock require a mounted volume.

…ems.

For better type safety. Also, renamed to separate the namespace.

…ties. This is done to distinguish the functions for checking the type of the filesystem from the functions for checking its properties. (Both regarding the original A. DOS filesystems.)

…unctions. Separating ADFlib's namespace, better type safety.

With static locals, the results of concurrent calls are unpredictible.

Separating ADFlib's namespace.

These arrays are in file header blocks and file ext. blocks. (Separating ADFlib's namespace).

Separating ADFlib's namespace.

Missed in b96349f, where it was renamed to adf_byteorder.

Some environments define NDEBUG already in release builds which causes a warning here.

This makes sure that it stays consistent with the .c file and keeps compilation and linking consistent too.

This was probably missed in a refactor.

The code seems to mix int, uint32 and SECNUM (which is int) for representing sector numbers. A furhter refactor may benefit from changing all those to SECNUM and makeing that unsigned unless something relies on sector numbers potentially being negative.

Warning fixes

Fixed failure case never tested

The comment is not relevant anymore (since a long time). The function Win32InitDevice does not have the 3rd parameter (at least since 0.7.12, which is the earliest in the repo...).

To simplify different ways of packaging, all packaging-related files should be stored in packaging/ directory.

Add preparing (copying) deb packaging configuration to required place. Add state validation (to prevent problems with execution).

Separating ADFlib's namespace.

…bel).

…locks). Not ignored mean the read operation will return failure in case of an invalid checksum (default).

Renaming the function to follow the naming convention: adfSubsystemFunction.

The only types to pass are bool and (function) pointers, both in practice ints. Simpler and more safe is to use intptr_t that passing by pointer (void *). (However, in case some more complex objects would be required, the previous way could be more flexible).

Separating ADFlib's namespace.

Renaming the function to follow the naming convention: adfSubsystemFunction.

Renaming the function as a typical setter.

Separating ADFlib's namespace.

(while searching for root block).

…sum errors'). Give a warning or error on an invalid checksum, depending on the env. setting.

No need to mention native dev. driver (it is the default).

Goals: 1. Separate expected output from scripts. 2. Make the tests work with MSYS2. 1. is required by 2. since under MSYS2 the output differs largely (EOL, paths). Mangling these in a script is maybe possible, but inherently problematic. The biggest problem comes from DOS/Windows-style paths using '\' as separator, which is also the escape character in shell scripts... Consequence: the escape characters are working somewhat differently on MSYS2 and on Linux and creating a common script is absurdly hard. Separating test outputs allows to use different files with expected contents depending on the platform. Note on unadf tests: file permission checks are not working on MSYS2 - either the way of checking or permissions are invalid. (File permission test is disabled in the uadf test script).

…f i/o operations.

vol->mounted remained set in case of error while reading root block - what lead to an invalid state of vol. Changed to unsetting it immediately after trying to read the root block. Also - setting it false by default (it was not initialized properly in case the filesystem was not present or was not a DOS filesystem (OFS/FFS).

…aw block data. Note that 'safe' concerns the size, not contents. The string can still contain invalid / binary data, which, for instance, if formatted in a way to perform an attack, can cause harm if outputted to a terminal. This has to be taken care in printing functions, as internally consistency with raw data from blocks must be kept.

Doing it directly with adfReadRDSKblock caused generating an error message, while there is no error (just the block not present). Now, the code first checks if the RDSK signature is present, then, if so, adfReadRDSKblock is called to load the block.

The device on which we operate can have the structure of a "hardfile", while now it will be a partitioned one, with RDSK block etc., so the device type must be set accordingly.

…ype. The device types HARDFILE and HARDDISK are used interchangably, depending if the device is partitioned (contains an RDSK block etc.) or not. Calling a create device function (which sets up the volume / partition structure), the type can be changed, so the mount function should be able to detect properly the type, and switch it both ways. The priority is changed to start from detecting the presence of an RDSK block, and, if it is not found, treat the device as a HARDFILE, an unpartitioned device, with a single volume using the whole device (like a floppy disk).

and rename the DEVTYPE_* consts to have the ADF_ prefix (to separate the ADFlib's namespace).

- Changed / improved command line options (with reasonable defaults). - Add possibility to 'enforce' formatting even if a filesystem already exists. - Allow formatting HDF volumes (hard disk-like large devices with floppy-like structure, ie. without the RDSK block, without partitions, just one volume on the whole device).

(For those platforms that do not have it).

The custom implementation of getopt does not allow to interleave the free, non-switch parameters. They must be provided at the end. (It would be good to improve that getopt implementation but this may happen later - not a priority...)

Checking the platform and presence of the function moved to build system (CMake or autotools). The C code checks only if the property is set.

…ation. This includes updating: tests, man pages, deb packaging and build system configuration.

…device. Also - break extracting in case of an error, otherwise the extraction code may go into an infinite loop.

Added info about development moved to the new repo.

block is allocated and left orphan untill it’s added to the list. If adfReadGenBlock() fails then block is never freed. The fix is an added call to free() in that case but also something to make sure that ownership of block is always transfered to the list so that it’s not freed twice. This also simplifies the final cleanup if().

This was caused by using a stack-based struct for block instead of allocating it. This meant adfFreeGenBlock() could not be called on it and therefore the name param was leaked.

Also: make const things const and update ret. code type name.

I think this is a bit overzealous from the Xcode analyzer but unless there is a reason to call adfReadRootBlock() where it’s currently called, it does receive a non-const pointer to vol and the code releis on vol->bitmap.size for a lot of its acesses to sectList. There is here nothing guarding against this value changing during the call to adfReadRootBlock(). This guarantees that the value used is constant throughout.

If rdsk.fileSysHdrList is -1 then fshd.segListBlock is never initialised to anything but it gets read before the following loop.

Testing for the two sums being equal is not the same thing as testing for each separate values being equal. This could lead to some values being un-initialized but also undetected by that test.

Error checks for all operations on adf device were added. The status returned by the program can be used to identify which operation failed.

This allows the filename to be changed without recompiling the test. Also, the test can be used with different files.

…ecover a parameter. This allows the file header block sector to be changed without recompiling the test (the test can be used with different files).

The undel3 test differed from undel2 only in 2 things: the adf device (OFS vs FFS) used and the file to delete/recover (its name and sector). After improving undel2 (prev. commits), merging them together became possible and an obvious improvement.

The status returned by the program can be used to identify which operation failed.

And return error if not.

And return a unique code allowing to identify it.

Add consistency checks. Allocate space for ext. blocks only if needed.

Also, make a common error handler (to reduce clutter).

This makes sure that other functions that passed blocksToRemove to us don't try to free it twice.

Asserts are removed in Release builds. This causes this safety check to be removed and potentially uninitilazed the values in blocksToRemove.sectors to be accessed.

Volume's fstype must be already set by adfDevMount or adfDevCreate. No need to read and parse the bootblock - besides filesystem signature, it is not needed for accessing filesystem data. Note that the requirement to load a valid bootblock is a problem in case of many non-bootable disks (even Workbench Extras), which do not have a valid bootblock, with correct checksums etc., but only the 'DOSn' signature at the beginning. Mounting such volumes was failing with a checksum error on reading bootblock, if 'ignoreChecksumErrors' was not set. This change alleviates the issue.

For some reason(?), it was disabled. Fixed.

… files. For now, the program allows only to display deleted entries.

…unction.

Also, use a better command-line option parser.

Adding prefix separating namespace and following the lib-ObjectTypeOrSubsys-Function scheme.

Also, rename the function for getting size in block without bootblock.

…n bitmap is marked valid.

Add vector item size. Add functions for allocating and releasing vector memory. Make changes where adfVector already used. The implementation chosen is made to be possibly simple and flexible, yet relatively safe and reliable. "Derived" vector types can be defined by creating struct type like the AdfVector, just with proper pointer type (the field can be renamed, like in AdfVectorSectors, but the order must be preserved!). It allows to write simple no-clutter code where it is used and gives type safety. The only moments when unsafe casts are needed are when calling adfVectorAllocate/Free, which must be done carefully (not to cast there any other than an AdfVector-like pointer!). Before calling addVectorAllocate, proper item size(!) and required number of items must be set. Then, the vector can be allocated/deallocated as needed (even without updating neither size, nor the number of items, as long as its structure exists).

…ne place. .itemSize set in main() was overwritten by memset -> fixed.

Also, fix memory leaks in adfUndelFile (in case of an error).

…sed). While files where undeleted successfully (visible, readable), there was an inconsistency in the volume's block allocation bitmap: the file header block of the recovered file was still marked as not used. The result was that the adf_salvage utility was still showing the file as 'deleted'. This would also result in inevitable corruption / data loss in case of allocating new blocks for writing - the file header block could be allocated (as a 'free' one) and overwritten with a block of another entry.

…ything...).

…n error.

Checking if the blocks of the file to undelete are marked 'free' was mixed with marking them 'used' by the file being undeleted. If, however, something failed, the changes had to be reverted. It is easier (and safer) to do all the checking before making any changes.

In case of an error on updating dir. cache during undeleting. (re)creating the file entry will result in more inconsistency and, therefore, it should be reverted.

Make clear, separated sections for dealing with entries having hash not yet present, and for entries with hash already present in the directory. A few other minor changes (declarations moved to narrowed scope, whitespaces).

The flags defined by AM_CPPFLAGS were not used in compilation due to invalid order, apparently AM_CFLAGS must be defined before AM_CPPFLAGS. More info: https://www.gnu.org/software/automake/manual/html_node/Flag-Variables-Ordering.html

Extracting the artifact requires permissions to write in /usr/local.

It must have been before as additional installation was not necessary (apparently some change in recent MacOs images on GH).

Installation requires permissions to write in /usr/local.

This commit fixes the error that started appearing on new MacOS images in GH actions. Apparently, due to some unknown change, the installed binaries had no longer the LC_RPATH set, what resulted in the error: -------------------------------------------------------- Run examples/tests/test_examples_basic.sh bin/ examples/tests/test_examples_basic.sh bin/ shell: /bin/bash -e {0} ============================================================= executing: unadf -r examples/tests/arccsh.adf dyld[3945]: Library not loaded: @rpath/libadf.1.dylib Referenced from: [...] /usr/local/bin/unadf Reason: no LC_RPATH's found [...] examples/tests/test_examples_basic.sh: line 39: [...] Error: Process completed with exit code 134. -------------------------------------------------------- The INSTALL_RPATH property had to be set in CMake configuration to fix the problem. More details in the linked issue and on: https://discourse.cmake.org/t/how-to-get-an-lc-rpath-and-rpath-prefix-on-a-dylib-on-macos/5540/7

The packaging is now provided as an option, in a subdirectory. Official Debian (or other) packaging should be done in a separated repository, as it may have its own versioning etc.

And some minor corrections in testing with autotools section.

Also, correct sources that were relying on that.

Remove unneeded include (not even exising anymore...).

2 unit test were using the same adf filenames as test devices, so the tests, when executed in parallel (building debs) were failing. Fixed making unique filenames.

Library headers listed as _HEADERS are being installed along with the library, while the private sources should not (they are needed only for building the library itself). Such private headers can be specified as noist_HEADERS, or, what is recommender by Automake manual, just as _SOURCES. Note, that if the private headers are completely omitted, make distcheck fails.

The DLL export prefix is required only in headers (in declarations).

Separating ADFlib's namespace.

The tested function is from a private library module, it is not exported by the library, so the module must be linked directly.

Lintian complained about privacy breach.

0.9.0

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

0.9.0 #81

0.9.0 #81

Commits on Feb 20, 2024

Commits on Feb 21, 2024

Commits on Feb 22, 2024

Commits on Feb 26, 2024

Commits on Feb 27, 2024

Commits on Feb 28, 2024

Commits on Feb 29, 2024

Commits on Mar 1, 2024

Commits on Mar 2, 2024

Commits on Mar 8, 2024

Commits on Mar 9, 2024

Commits on Mar 12, 2024

Commits on Mar 13, 2024

Commits on Mar 14, 2024

Commits on Mar 16, 2024

Commits on Mar 17, 2024

Commits on Mar 18, 2024

Commits on Mar 19, 2024

Commits on Mar 20, 2024

Commits on Mar 21, 2024

Commits on Mar 22, 2024

Commits on Mar 26, 2024

Commits on Mar 27, 2024

Commits on Mar 28, 2024

Commits on May 13, 2024

Commits on May 14, 2024

Commits on May 15, 2024

Commits on May 16, 2024

0.9.0 #81

Are you sure you want to change the base?

0.9.0 #81

Commits on Feb 20, 2024

Commits on Feb 21, 2024

Commits on Feb 22, 2024

Commits on Feb 26, 2024

Commits on Feb 27, 2024

Commits on Feb 28, 2024

Commits on Feb 29, 2024

Commits on Mar 1, 2024

Commits on Mar 2, 2024

Commits on Mar 8, 2024

Commits on Mar 9, 2024

Commits on Mar 12, 2024

Commits on Mar 13, 2024

Commits on Mar 14, 2024

Commits on Mar 16, 2024

Commits on Mar 17, 2024

Commits on Mar 18, 2024

Commits on Mar 19, 2024

Commits on Mar 20, 2024

Commits on Mar 21, 2024

Commits on Mar 22, 2024

Commits on Mar 26, 2024

Commits on Mar 27, 2024

Commits on Mar 28, 2024

Commits on May 13, 2024

Commits on May 14, 2024

Commits on May 15, 2024

Commits on May 16, 2024