Fetch and download

fetch() — a GET into memory

A guarded GET whose body is read into memory, within a byte cap and one time budget:

$page = SafeFetch::fetch(
    $url,
    maxBytes: 1_000_000,          // default safe-fetch.fetch.max_bytes (5 MiB)
    timeout: 20,                  // default safe-fetch.fetch.timeout (30 s), every hop together, 1–600
    accept: 'text/html',
    headers: ['X-Trace' => $id],  // never Host; sent to the first origin only
    followRedirects: true,        // default: the policy's redirects
    policy: SafeFetch::policy(),  // default: the configured policy
);

$page->status;       // 2xx
$page->body;         // the whole body, within the cap
$page->contentType;  // or null
$page->url;          // the URL the answer came from (the last hop, rebuilt)
$page->headers;      // array<string, list<string>>
$page->header('etag');
$page->json('data.items', []);  // the body decoded as JSON (null when it is not), a dot-notation key

FetchedResponse::json(?string $key = null, mixed $default = null) decodes the body on every call (objects as arrays); a body that is not valid JSON is null, so a key reads $default.

download() — the same GET, streamed to a file

For a file too large to hold in memory — an import, a media upload by URL. It takes the same options as fetch() and fails the same way; the body goes through the same cap into the file, never into memory:

$file = SafeFetch::download($url, $path, maxBytes: 50_000_000);   // $path: e.g. storage_path('app/imports/feed.xml')

$file->status;       // 2xx
$file->path;         // where the body is — the path you gave
$file->bytes;        // its size, as written (decoded)
$file->contentType;  // or null
$file->url;          // the URL the answer came from (the last hop, rebuilt)
$file->headers;      // array<string, list<string>>
$file->header('etag');

• The directory must exist and be writable, and the path must not be a read-only file or a directory — otherwise InvalidArgumentException, before anything is sent.
• The body goes to a temporary file next to $path (.<name>.<random>) and is renamed over $path only when the download succeeded — an atomic swap on one filesystem. A file already at $path stays whole meanwhile and keeps its file mode; a new file gets the umask’s.
• A symlink at $path is written through, as a plain write would: the temporary file sits next to the link’s target and replaces that, and the link stays a link.
• Every hop starts the temporary file afresh, so a redirect’s body never lands in it.
• On any failure the temporary file is deleted — a body over the cap, a non-2xx status, a refused redirect, a dropped connection, a full disk — and a file already at $path is left as it was. A rename that fails throws RuntimeException.
• If the worker dies hard mid-download, the temporary .<name>.XXXXXX file can remain in that directory — safe to delete; nothing reads it.

Behaviour of both

• Only a 2xx body is read. Any other answer is judged by its status and headers alone — the transfer stops before its body, so a 503 with a huge error page stays a retryable 503. An informational answer (103 Early Hints) is passed over.
• The byte cap: a 2xx declaring a Content-Length over the cap stops before its body; any other body is cut off as it passes the cap — counted after decoding, so a compressed bomb stops too. Either way: FetchFailed (too_large).
• Redirects: followRedirects: true (or the policy) follows all five statuses for a GET unless the policy names its own; only the final body is kept.
• Headers you pass go to the first origin only, and a Host header is refused (UnsafeRequestOptions).
• The request comes from the container’s HTTP client factory, so Http::fake() answers and records it.

Failures

use RoundlyConsulting\SafeFetch\Exceptions\FetchFailed;
use RoundlyConsulting\SafeFetch\Exceptions\SafeFetchException;

try {
    $page = SafeFetch::fetch($url, maxBytes: 1_000_000);
} catch (FetchFailed $e) {
    $e->reason();       // Reason::Status, TooLarge, Timeout, Connection, Tls or Protocol
    $e->status();       // the HTTP status for Reason::Status, otherwise null
    $e->retryAfter();   // the server's Retry-After in seconds, for 408 / 429 / 5xx
    $e->retryable();
} catch (SafeFetchException $e) {
    $e->retryable();    // DestinationUnavailable (unresolved, timeout): yes · DestinationRejected: never
}

• Retry-After is read as seconds or an HTTP date, capped at 86 400.
• A transport failure’s getPrevious() is a RuntimeException naming only the underlying exception’s class and curl code — never Guzzle’s own message, which names the URL and the address it connected to.
• Only transport failures are mapped. Anything else reaches the caller as thrown — a refusal of the guard, the fake’s LogicException for a request no Http::fake() stub answered, Http::preventStrayRequests()’s StrayRequestException, a programmer error. download() then leaves a file already at the path as it was.