The `max_iterations` value in `#parse_curl_output` is only intended
to prevent its `while` loop from potentially turning into an endless
loop. This should only come into play in exceptional circumstances
but the current default value (5) is low enough that we're seeing it
under normal circumstances.
`#parse_curl_output` isn't intended to restrict the number of
redirections (this should be done using the `--max-redirs` option in
`curl) but it's effectively doing this in rare cases due to the low
`max_iterations` default. This is a problem because `curl` can
successfully return a response only to have `#parse_curl_output`
error in relation to `max_iterations`.
Originally the code in `#parse_curl_output` was used in the context
of livecheck, where it's not a huge issue if a check fails. However,
now the `#parse_curl_output` method is used in important parts of
brew like `#curl_download`. We've received a report of a download
failing with the "Too many redirects (max = 5)` error, effectively
preventing the user from installing a formula [from a third-party
tap].
Until we can come up with a more adaptive way of bounding this
`while` loop, I think we should simply raise the default to something
that's less likely to be encountered under normal circumstances
(e.g., 25).
Current situation with update.sh & update-reset.sh:
Upon creating a new tap and before adding remote 'origin' to it,
running `brew update` produces multiple cryptic error messages.
This change converts these cryptic Git messages into clear warnings
about absent remote 'origin'.
How to test:
1. Create a new tap using `brew tap-new`
2. Run `brew update`
The `#curl_http_content_headers_and_checksum` method previously
parsed responses from `curl` output even if `status.success?` wasn't
`true`. A recent commit of mine moved the parsing logic behind this
guard but it's now leading to a "...is not reachable" error when a URL
involves a large download that takes longer than 25 seconds to finish
and hits the timeout.
This commit resolves the issue for the time being by moving related
logic back to its previous location, where it isn't guarded by
`status.success?`.
Results in searching a list of tokens from official Homebrew cask taps and fully-qualified tokens from other taps, matching the list when searching formulae.
When its `try_partial` argument is `true`, `#curl_download` makes a
`HEAD` request before downloading the file using `#curl`. Currently
`try_partial` defaults to `true`, so any `#curl_download` call that
doesn't explicitly specify `try_partial: false` will make a `HEAD`
request first. This can potentially involve several requests if the
URL redirects, so it can be a bit of unnecessary overhead when a
partial download isn't needed.
Partial downloads are generally only useful when we're working with
larger files, however there's currently only one place in brew where
`#curl_download` is used and this is the case:
`CurlDownloadStrategy`. The other `#curl_download` calls are fetching
smaller [text] files and don't need to support partial downloads.
This commit changes the default `try_partial` value to `false`,
making partial downloads opt-in rather than opt-out.
We want `try_partial` to continue to default to `true` in
`CurlDownloadStrategy` and there are various ways to accomplish this.
In this commit, I've chosen to update its `#initialize` method to
accept a `try_partial` argument that defaults to `true`, as this
value can also be used in classes that inherit from
`CurlDownloadStrategy` (e.g., `HomebrewCurlDownloadStrategy`). This
instance variable is passed to `#curl_download` in related methods,
effectively maintaining the previous `try_partial: true` value, while
also allowing this value to be overridden when necessary.
Other uses of `#curl_download` in brew are
`Formulary::FromUrlLoader#load_file` and
`Cask::CaskLoader::FromURILoader#load`, which did not provide a
`try_partial` argument but should have been using
`try_partial: false`. With the `try_partial: false` default in this
commit, these calls are now fine without a `try_partial` argument.
The only other use of `#curl_download` in brew is
`SPDX#download_latest_license_data!`. These calls were previously
using `try_partial: false` but we can now omit this argument with
the new `false` default (aligning with the above).
The default redirection maximum for `curl` is 50 but we should use
something more reasonable in livecheck. It's rare but a misconfigured
server with an endless redirection loop will hit the 50 redirection
limit. Unfortunately, we've encountered this in the wild (e.g., the
server for `getmail` and `memtester` endlessly redirects), so it's
not an idle concern. This commit basically adds `--max-redirs 5` to
`Livecheck::Strategy::DEFAULT_CURL_ARGS` to enforce a more reasonable
redirection maximum.
To be clear, the `max_iterations` logic in `#parse_curl_output`
(which was previously found in `Strategy#page_content`) doesn't
restrict the number of redirections that `curl` follows. At the point
the `curl` output is being parsed, the requests have already been
made and `max_iterations` simply restricts the number of responses
`#parse_curl_output` is willing to parse. If we use `--max-redirs`
and properly set `max_iterations` to `max-redirs + 1`, we shouldn't
encounter the "Too many redirects" error in `#parse_curl_output`.
In cases where there may be more than five responses in `curl`
output to parse, we need to be able to control the `max_iterations`
of the `while` loop in `#parse_curl_output` to properly parse all
the responses.
For example, if we pass `--max-redirs 5` to `curl` and there are
exactly five redirections before the final response, the output would
contain a total of six responses and `#parse_curl_output` wouldn't
properly handle this (it would give a `Too many redirects` error).
`max_iterations` should be the maximum number of redirections + 1
(to account for any final response after the redirections), so we
need to be able to override this value when necessary.
Mike McQuaid