ntruessel / etar Goto Github PK

Giving an invalid path to archive returns an operating system error

The ustar utilities should contain links to the tar specification

Some of the tests (most of the tests that create/open files) have machine dependent metadata. Check whether we can find a way to fix this

All calculations/writing and so forth in ARCHIVABLE and FILE_ARCHIVABLE assume that USTAR_HEADER_WRITER is used

The UNARCHIVER parent class has slightly changed

ATM USTAR_HEADER_PARSER does error reporting by setting last_parsed_header to Void. As proposed by @jocelyn in #23 this could be improved by e.g. introducing an error_message attribute that indicates where the first error occurred.

ATM files in FILE_ARCHIVABLE are duplicated with the .twin feature. Maybe this is not enough to create a copy like it is intended. To be tested (and fixed if necessary).

PAX provides a reserved identifier for mtime whereas USTAR_HEADER_WRITER does not even check whether the value might be too large.

Either argument why this check can be omitted or implement it

Pax global headers are not tested at all

when trying to extract foo.tar ... I do

minipax -r -f foo.tar

but if foo.tar does not exists ... the execution violates assertions, it is missing a simple file existence checking .. and appropriate error message.

This is minor issue for the example, but could be annoying for a tool.

Some classes use make, some use default_create. Should be unified

Just found another uncovered feature with a bug in its precondition.

PAX_UNARCHIVER does not parse its entries cleanly, e.g. a space in some key/value will cause the parser to throw an error

Both USTAR_HEADER_PARSER and USTAR_HEADER_WRITER calculate the checksum using own features. These should be merged (e.g. in some TAR_UTILS parent class?)

As written in #9, verify_checksum needs performance improvements

ATM this feature adds one block for the header which might be correct for the ustar format but is wrong for pax (it is dependent on the header writer)

ATM the FILE_UNARCHIVER only creates a new file with the correct filename, but without setting any metadata that the header provides (like user, group, mode, mtime...)

The constants in TAR_CONST (mostly offset and length) should be grouped in such a way that mistakes like

parse_string (block, {TAR_CONST}.tar_header_devmajor_offset, {TAR_CONST}.tar_header_devminor_length) 

become more obvious

This could be done by creating a custom class (+ once routine) for them (but then, the advantage of not having to inherit from TAR_CONST is lost)

The method USTAR_HEADER_WRITER is quite inefficient: It initializes Result as an empty string which is inefficient most of the cases. There is still room for improvements (parse_string is one of the core features used by the header parsers).

A header parser that does not have any error messages should always have an attached header (ATM the header can be Void even when no error messages were logged)

ATM error handling is rather cumbersome. For a single class it is quite easy, but as soon as this class has attributes that can produce errors as well it gets cumbersome to check whether an error occurred (have to check all attributes).

I propose some interface that allows me to register an error logger/reporter for classes that need it, such that errors can be reported / checked in a central place.

The eiffel documentation tells that path strings are required to use the "escaped representation as described in UTF_CONVERTER".

Have to check what exactly this means and whether the library uses it / how to change it to do it correctly

On windows, set_date for files which are a directory fails (Operating System Error)

The code that calculates the needed blocks exists twice, once for ARCHIVABLE, once for UNARCHIVER. See #14

For the last entry, pax unarchiver does not check whether the parsing state machine is in an accepting state

The ustar header parsing is so slow that the content listing demo takes ages to list the contents of an archive for which bsdtar needs milliseconds (archives/etar.tar). This is possibly caused by

1. The content listing demo not skipping blocks that should be skipped
2. No early termination in the ustar header parsing feature on parse failure
3. Slow string parsing feature in TAR_HEADER_PARSER

The test set for ustar headers does not contain any tests that use file names with non-ASCII characters. There should be tests for that as well

Currently only complete writing is tested, all block-wise writing features are untested.

ATM PAX_HEADER_PARSER does not check whether a pax header is global or extended. Therefore it will probably crash if one feeds it with a global header followed by an extended header, followed by an ustar header

The mtime format in extended headers seems to differ from the header mtime format. However we do write mtime in this special format yet

Nearly all classes that report errors are not tested for cases where errors occur

ATM all tests check whether valid inputs are accepted. There should also be tests that check whether invalid inputs are rejected.

Paths containing .. and absolute paths should be disabled

Setting uid/gid to 0 causes an "Operation not permitted" operating system error.

To demonstrate some parts of the library, I want an example that outputs all the contents of a tar file, but not their payload.