`check-generated-files -u` aborted if one of the generated files
didn't exist. Now it treats a missing file as an out-of-date file.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Changelog contents should be UTF-8 text files. So explicitly open all files as
UTF-8. This makes the script independent of the ambient locale (except with
respect to exception messages, but we can live with that).
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Changelog contents should be UTF-8 text files. There's no need to be
binary-safe. So switch to using text strings in Python (str, not bytes). This
commit makes the following changes:
* Bytes literals (b'…') to string literals ('…').
* Subprocess output (which is all git information) is decoded as ascii.
* Inject text directly in exceptions rather than calling a decode method.
This is enough to make the script work as desired in a UTF-8 locale.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
"exit /b" without an explicit exit value doesn't copy the value of the
last command executed, causing issues on Jenkins.
Signed-off-by: Bence Szépkúti <bence.szepkuti@arm.com>
Accept Windows line endings in input files on any platform. This makes
the scripts work even when running a Unix perl with a source tree that
has Windows line endings, as happens for example on our Travis Windows
instances. This change is harmless in the common case where the input
has the platform's default line endings.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Instead of manipulating CR explicitly to cope with CRLF (Windows) line
endings in input and produce output with CRLF line endings, just
convert files from/to CRLF line endings when reading/writing.
The minimum required Perl version remains 5.8, since this both the
version that introduced Digest::MD5 (which was used before this patch)
and the version that introduced open "<:crlf" (which this patch
introduces).
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Don't run `make list` to obtain the list of programs in
generate_visualc_files.pl. This doesn't work on Windows when a `make`
command is not available. Instead, read the makefile.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Don't regenerate configuration-independent files when config.h or
crypto_config.h changes. These files only depend on the set of symbols
present in the headers and not on which symbols are enabled. To avoid
rebuilding the generated files whenever the configuration changes,
don't declare the configuration as a dependency.
In the rare event that a maintainer makes an edit to *config.h that
affects the generated files, they'll have to remove the generated files.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
On Linux, all.sh takes care of it. On Windows, run
make_generated_files.bat.
Perl is now required on Windows.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Now that generated source files are no longer checked in version
control, they must be generated before running any tests.
Do not check the generated files for freshness: it's no longer relevant.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
The main benefit is that this ensures that the directory always exists in a
Git checkout. This way the maintenance scripts don't have to worry about the
case where the directory doesn't exist.
Also it unclutters `/.gitignore`.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
Conflicts:
* configs/config-psa-crypto.h: modified here, removed in development
* tests/suites/test_suite_x509parse.data: all conflicts are in depends_on
lines where development made a change unrelated to MBEDTLS_SHAxxx and our
branch either changed `MBEDTLS_SHA256_C` to `MBEDTLS_SHA224_C` or
`MBEDTLS_SHA512_C:!MBEDTLS_SHA512_NO_SHA384` to ``MBEDTLS_SHA384_C`, with
no change to what the test does. Pick the other branch's dependency
changes then apply our SHA dpeendency change.
To simplify the documentation, and because there are only two realistic
behaviors in practice, only allow two behaviors for multipart output:
immediate output, or buffering of the last partial block.
State some simple rules that applications can follow if they don't care about
the details.
Explicitly state how much output is needed for finish(). Only require the
buffer size to be the size of the actual output, not the size of the potential
output in the worst case. Rename the parameter from output_len to output_size
since it's a buffer size and not necessarily the length of the data.
No longer guarantee that the built-in implementation produces immediate
output.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
