diff options
| author | Tatsuhiro Tsujikawa <tatsuhiro.t@gmail.com> | 2015-12-27 18:39:47 +0900 |
|---|---|---|
| committer | Tatsuhiro Tsujikawa <tatsuhiro.t@gmail.com> | 2015-12-27 18:40:08 +0900 |
| commit | b1132d6b10cd3791b9529d7031d723fd0e2d8215 (patch) | |
| tree | 9ce465b42bf09372118335c7c619bb2b41869eaf /src/HttpResponse.cc | |
| parent | 4abad2f64c9ff80144a61c5a9c2bea144852b883 (diff) | |
make clang-format using clang-format-3.6
Diffstat (limited to 'src/HttpResponse.cc')
| -rw-r--r-- | src/HttpResponse.cc | 105 |
1 files changed, 47 insertions, 58 deletions
diff --git a/src/HttpResponse.cc b/src/HttpResponse.cc index 02f4b134..f009ecae 100644 --- a/src/HttpResponse.cc +++ b/src/HttpResponse.cc @@ -61,33 +61,29 @@ #include "array_fun.h" #include "MessageDigest.h" #ifdef HAVE_ZLIB -# include "GZipDecodingStreamFilter.h" +#include "GZipDecodingStreamFilter.h" #endif // HAVE_ZLIB namespace aria2 { -HttpResponse::HttpResponse() - : cuid_{0} -{} +HttpResponse::HttpResponse() : cuid_{0} {} void HttpResponse::validateResponse() const { int statusCode = getStatusCode(); - switch(statusCode) { + switch (statusCode) { case 200: // OK case 206: // Partial Content if (!httpHeader_->defined(HttpHeader::TRANSFER_ENCODING)) { // compare the received range against the requested range auto responseRange = httpHeader_->getRange(); if (!httpRequest_->isRangeSatisfied(responseRange)) { - throw DL_ABORT_EX2(fmt(EX_INVALID_RANGE_HEADER, - httpRequest_->getStartByte(), - httpRequest_->getEndByte(), - httpRequest_->getEntityLength(), - responseRange.startByte, - responseRange.endByte, - responseRange.entityLength), - error_code::CANNOT_RESUME); + throw DL_ABORT_EX2( + fmt(EX_INVALID_RANGE_HEADER, httpRequest_->getStartByte(), + httpRequest_->getEndByte(), httpRequest_->getEntityLength(), + responseRange.startByte, responseRange.endByte, + responseRange.entityLength), + error_code::CANNOT_RESUME); } } return; @@ -118,8 +114,8 @@ void HttpResponse::validateResponse() const std::string HttpResponse::determineFilename() const { - std::string contentDisposition = util::getContentDispositionFilename - (httpHeader_->find(HttpHeader::CONTENT_DISPOSITION)); + std::string contentDisposition = util::getContentDispositionFilename( + httpHeader_->find(HttpHeader::CONTENT_DISPOSITION)); if (contentDisposition.empty()) { auto file = httpRequest_->getFile(); file = util::percentDecode(file.begin(), file.end()); @@ -129,9 +125,8 @@ std::string HttpResponse::determineFilename() const return file; } - A2_LOG_INFO(fmt(MSG_CONTENT_DISPOSITION_DETECTED, - cuid_, - contentDisposition.c_str())); + A2_LOG_INFO( + fmt(MSG_CONTENT_DISPOSITION_DETECTED, cuid_, contentDisposition.c_str())); return contentDisposition; } @@ -140,17 +135,15 @@ void HttpResponse::retrieveCookie() Time now; auto r = httpHeader_->equalRange(HttpHeader::SET_COOKIE); for (; r.first != r.second; ++r.first) { - httpRequest_->getCookieStorage()->parseAndStore - ((*r.first).second, - httpRequest_->getHost(), - httpRequest_->getDir(), - now.getTimeFromEpoch()); + httpRequest_->getCookieStorage()->parseAndStore( + (*r.first).second, httpRequest_->getHost(), httpRequest_->getDir(), + now.getTimeFromEpoch()); } } bool HttpResponse::isRedirect() const { - switch(getStatusCode()) { + switch (getStatusCode()) { case 300: // Multiple Choices case 301: // Moved Permanently case 302: // Found @@ -166,14 +159,12 @@ void HttpResponse::processRedirect() { const auto& req = httpRequest_->getRequest(); if (!req->redirectUri(util::percentEncodeMini(getRedirectURI()))) { - throw DL_RETRY_EX(fmt("CUID#%" PRId64 - " - Redirect to %s failed. It may not be a valid URI.", - cuid_, - req->getCurrentUri().c_str())); + throw DL_RETRY_EX(fmt( + "CUID#%" PRId64 " - Redirect to %s failed. It may not be a valid URI.", + cuid_, req->getCurrentUri().c_str())); } - A2_LOG_INFO(fmt(MSG_REDIRECT, - cuid_, + A2_LOG_INFO(fmt(MSG_REDIRECT, cuid_, httpRequest_->getRequest()->getCurrentUri().c_str())); } @@ -220,8 +211,8 @@ std::unique_ptr<StreamFilter> HttpResponse::getContentEncodingStreamFilter() const { #ifdef HAVE_ZLIB - if(util::strieq(getContentEncoding(), "gzip") || - util::strieq(getContentEncoding(), "deflate")) { + if (util::strieq(getContentEncoding(), "gzip") || + util::strieq(getContentEncoding(), "deflate")) { return make_unique<GZipDecodingStreamFilter>(); } #endif // HAVE_ZLIB @@ -231,7 +222,7 @@ HttpResponse::getContentEncodingStreamFilter() const int64_t HttpResponse::getContentLength() const { - if(!httpHeader_) { + if (!httpHeader_) { return 0; } @@ -240,7 +231,7 @@ int64_t HttpResponse::getContentLength() const int64_t HttpResponse::getEntityLength() const { - if(!httpHeader_) { + if (!httpHeader_) { return 0; } @@ -249,7 +240,7 @@ int64_t HttpResponse::getEntityLength() const std::string HttpResponse::getContentType() const { - if(!httpHeader_) { + if (!httpHeader_) { return A2STR::NIL; } @@ -274,10 +265,7 @@ void HttpResponse::setHttpRequest(std::unique_ptr<HttpRequest> httpRequest) httpRequest_ = std::move(httpRequest); } -int HttpResponse::getStatusCode() const -{ - return httpHeader_->getStatusCode(); -} +int HttpResponse::getStatusCode() const { return httpHeader_->getStatusCode(); } Time HttpResponse::getLastModifiedTime() const { @@ -299,11 +287,11 @@ bool parseMetalinkHttpLink(MetalinkHttpEntry& result, const std::string& s) } auto last = std::find(first, s.end(), '>'); - if(last == s.end()) { + if (last == s.end()) { return false; } - auto p = util::stripIter(first+1, last); + auto p = util::stripIter(first + 1, last); if (p.first == p.second) { return false; } @@ -323,17 +311,18 @@ bool parseMetalinkHttpLink(MetalinkHttpEntry& result, const std::string& s) break; } - if(value.empty()) { - if(name == "pref") { + if (value.empty()) { + if (name == "pref") { result.pref = true; } continue; } - if(name == "rel") { - if(value == "duplicate") { + if (name == "rel") { + if (value == "duplicate") { ok = true; - } else { + } + else { ok = false; } continue; @@ -341,8 +330,8 @@ bool parseMetalinkHttpLink(MetalinkHttpEntry& result, const std::string& s) if (name == "pri") { int32_t priValue; - if(util::parseIntNoThrow(priValue, value)) { - if(1 <= priValue && priValue <= 999999) { + if (util::parseIntNoThrow(priValue, value)) { + if (1 <= priValue && priValue <= 999999) { result.pri = priValue; } } @@ -362,14 +351,14 @@ bool parseMetalinkHttpLink(MetalinkHttpEntry& result, const std::string& s) // Metalink/HTTP is defined by http://tools.ietf.org/html/rfc6249. // Link header field is defined by http://tools.ietf.org/html/rfc5988. -void HttpResponse::getMetalinKHttpEntries -(std::vector<MetalinkHttpEntry>& result, - const std::shared_ptr<Option>& option) const +void HttpResponse::getMetalinKHttpEntries( + std::vector<MetalinkHttpEntry>& result, + const std::shared_ptr<Option>& option) const { auto p = httpHeader_->equalRange(HttpHeader::LINK); for (; p.first != p.second; ++p.first) { MetalinkHttpEntry e; - if(parseMetalinkHttpLink(e, (*p.first).second)) { + if (parseMetalinkHttpLink(e, (*p.first).second)) { result.push_back(e); } } @@ -379,12 +368,12 @@ void HttpResponse::getMetalinKHttpEntries if (option->defined(PREF_METALINK_LOCATION)) { const std::string& loc = option->get(PREF_METALINK_LOCATION); util::split(loc.begin(), loc.end(), std::back_inserter(locs), ',', true); - for (auto& l: locs) { + for (auto& l : locs) { util::lowercase(l); } } - for (auto& r: result) { - if(std::find(locs.begin(), locs.end(), r.geo) != locs.end()) { + for (auto& r : result) { + if (std::find(locs.begin(), locs.end(), r.geo) != locs.end()) { r.pri -= 999999; } } @@ -412,7 +401,7 @@ void HttpResponse::getDigest(std::vector<Checksum>& result) const util::lowercase(hashType); digest = base64::decode(digest.begin(), digest.end()); if (!MessageDigest::supports(hashType) || - MessageDigest::getDigestLength(hashType) != digest.size()) { + MessageDigest::getDigestLength(hashType) != digest.size()) { continue; } @@ -425,8 +414,8 @@ void HttpResponse::getDigest(std::vector<Checksum>& result) const for (auto i = result.begin(), eoi = result.end(); i != eoi;) { bool ok = true; auto j = i + 1; - for(; j != eoi; ++j) { - if((*i).getHashType() != (*j).getHashType()) { + for (; j != eoi; ++j) { + if ((*i).getHashType() != (*j).getHashType()) { break; } |
