Re: [RFC PATCH] Add new build target 'check-spelling'

[Thomas Huth]

On 31/10/2022 11.44, Stefan Weil wrote:
> Am 31.10.22 um 08:52 schrieb Thomas Huth:
>
> > On 31/10/2022 08.43, Stefan Weil wrote:
> > > `make check-spelling` can now be used to get a list of spelling errors.
> > > It uses the latest version of codespell, a spell checker implemented in 
> > > Python.
> > > Signed-off-by: Stefan Weil <sw@weilnetz.de>
> > > ---
> > >
> > > This RFC can already be used for manual tests, but still reports false
> > > positives, mostly because some variable names are interpreted as words.
> > > These words can either be ignored in the check, or in some cases the code
> > > might be changed to use different variable names.
> > >
> > > The check currently only skips a few directories and files, so for example
> > > checked out submodules are also checked.
> > >
> > > The rule can be extended to allow user provided ignore and skip lists,
> > > for example by introducing Makefile variables CODESPELL_SKIP=userfile
> > > or CODESPELL_IGNORE=userfile. A limited check could be implemented by
> > > providing a base directory CODESPELL_START=basedirectory, for example
> > > CODESPELL_START=docs.
> > >
> > > Regards,
> > > Stefan
> [...]
> > > I like the idea, but I think it's unlikely that we can make this work for 
> > > the whole source tree any time soon. So maybe it makes more sense to 
> > > start with some few directories first (e.g. docs/ ) and then the 
> > > maintainers can opt-in by cleaning up their directories first and then by 
> > > adding their directories to this target here?
> >  Thomas
>
> Even without implementing CODESPELL_START as described above, the script can 
> already be used and integrated into CI scripts.
> It takes about 60 seconds to check the whole source tree including 
> submodules on my (slow) virtual machine.
> The resulting output has about 20000 lines or 1272 KiB. It can be filtered 
> for relevant parts of the source tree or used for a summary.
> Sample script: grep "^[.]" spellcheck.log | sed s/^..// | sed 's/\/.*//' | 
> sed s/:.*// | sort | uniq -c
> This produces a summary for the top level hierarchy of files and directories:
>
>        3 accel
>        1 audio
>        1 backends
>       77 block
>        7 block.c
>       20 bsd-user
>      386 capstone
>       12 chardev
>        1 configure
>        8 contrib
>        6 crypto
>       64 disas
>       32 docs
>       31 dtc
>        8 fpu
>        1 gdbstub
>        1 gdb-xml
>        1 .github
>      537 hw
>        7 inc
>      114 include
>        1 libdecnumber
>       33 linux-user
>        1 MAINTAINERS
>      150 meson
>        6 meson.build
>       16 migration
>        1 nbd
>        5 net
>       12 pc-bios
>        7 python
>        3 qapi
>        2 qemu
>        5 qemu-options.hx
>       22 qga
>    14175 roms
>       43 scripts
>        3 semihosting
>       18 slirp
>        2 softmmu
>       59 subprojects
>      504 target
>        6 tcg
>        3 test.rb
>      175 tests
>        6 tools
>       20 ui
>        8 util
>
> It shows that "roms" contributes by far the most typos. Omitting it would 
> reduce the required time to 22 seconds and the number of typos found (2947 
> lines in output) very much.
"roms" mostly consists of third-party submodules that we do not have direct 
control of. I think this should definitely be omitted.
> "capstone" (which has no entry in MAINTAINERS)
That's likely because it has been a submodule that has been removed a while 
ago. "rm -rf capstone" should solve that issue on your local buildtree ;-)
(yes, that's another nuisance of submodules - the checked out files don't go 
away when the submodule gets removed)
 Thomas