Fossil SCM
Based on discussions in [forum:f60dece061c364d1|forum thread f60dece061c364d1], (A) re-add the charset=utf-8 for text/* mimetypes, (B) extend the set of gzip-compressible mimetypes (e.g. JSON, wasm, tcl, tar), and (C) refactor (B)'s impl so that adding new types does not add a performance hit (it's faster now for most mimetypes).
Commit
86db2d94c60e23845362e5793280583223405f79c505f834c8aa13a60db9d9c5
Parent
ba1be566ae221a9…
1 file changed
+47
-4
+47
-4
| --- src/cgi.c | ||
| +++ src/cgi.c | ||
| @@ -330,13 +330,36 @@ | ||
| 330 | 330 | ** Return true if the response should be sent with Content-Encoding: gzip. |
| 331 | 331 | */ |
| 332 | 332 | static int is_gzippable(void){ |
| 333 | 333 | if( g.fNoHttpCompress ) return 0; |
| 334 | 334 | if( strstr(PD("HTTP_ACCEPT_ENCODING", ""), "gzip")==0 ) return 0; |
| 335 | - return strncmp(zContentType, "text/", 5)==0 | |
| 336 | - || sqlite3_strglob("application/*xml", zContentType)==0 | |
| 337 | - || sqlite3_strglob("application/*javascript", zContentType)==0; | |
| 335 | + /* Maintenance note: this oddball structure is intended to make | |
| 336 | + ** adding new mimetypes to this list less of a performance hit than | |
| 337 | + ** doing a strcmp/glob over a growing set of compressible types. */ | |
| 338 | + switch(zContentType ? *zContentType : 0){ | |
| 339 | + case (int)'a': | |
| 340 | + if(0==strncmp("application/",zContentType,12)){ | |
| 341 | + const char * z = &zContentType[12]; | |
| 342 | + switch(*z){ | |
| 343 | + case (int)'j': | |
| 344 | + return fossil_strcmp("javascript", z)==0 | |
| 345 | + || fossil_strcmp("json", z)==0; | |
| 346 | + case (int)'w': return fossil_strcmp("wasm", z)==0; | |
| 347 | + case (int)'x': | |
| 348 | + return fossil_strcmp("x-tcl", z)==0 | |
| 349 | + || fossil_strcmp("x-tar", z)==0; | |
| 350 | + default: | |
| 351 | + return sqlite3_strglob("*xml", z)==0; | |
| 352 | + } | |
| 353 | + } | |
| 354 | + break; | |
| 355 | + case (int)'i': | |
| 356 | + return fossil_strcmp(zContentType, "image/svg+xml")==0; | |
| 357 | + case (int)'t': | |
| 358 | + return fossil_strncmp(zContentType, "text/", 5)==0; | |
| 359 | + } | |
| 360 | + return 0; | |
| 338 | 361 | } |
| 339 | 362 | |
| 340 | 363 | |
| 341 | 364 | /* |
| 342 | 365 | ** The following routines read or write content from/to the wire for |
| @@ -419,10 +442,29 @@ | ||
| 419 | 442 | if( !g.httpUseSSL ){ |
| 420 | 443 | fflush(g.httpOut); |
| 421 | 444 | } |
| 422 | 445 | } |
| 423 | 446 | |
| 447 | +/* | |
| 448 | +** Given a Content-Type value, returns a string suitable for appending | |
| 449 | +** to the Content-Type header for adding (or not) the "; charset=..." | |
| 450 | +** part. It returns an empty string for most types or if zContentType | |
| 451 | +** is NULL. | |
| 452 | +** | |
| 453 | +** See forum post f60dece061c364d1 for the discussions which lead to | |
| 454 | +** this. Previously we always appended the charset, but WASM loaders | |
| 455 | +** are pedantic and refuse to load any responses which have a | |
| 456 | +** charset. Also, adding a charset is not strictly appropriate for | |
| 457 | +** most types (and not required for many others which may ostensibly | |
| 458 | +** benefit from one, as detailed in that forum post). | |
| 459 | +*/ | |
| 460 | +static const char * content_type_charset(const char *zContentType){ | |
| 461 | + if(zContentType!=0){ | |
| 462 | + if(0==strncmp(zContentType,"text/",5)) return "; charset=utf-8"; | |
| 463 | + } | |
| 464 | + return ""; | |
| 465 | +} | |
| 424 | 466 | |
| 425 | 467 | /* |
| 426 | 468 | ** Generate the reply to a web request. The output might be an |
| 427 | 469 | ** full HTTP response, or a CGI response, depending on how things have |
| 428 | 470 | ** be set up. |
| @@ -493,11 +535,12 @@ | ||
| 493 | 535 | |
| 494 | 536 | /* Content intended for logged in users should only be cached in |
| 495 | 537 | ** the browser, not some shared location. |
| 496 | 538 | */ |
| 497 | 539 | if( iReplyStatus!=304 ) { |
| 498 | - blob_appendf(&hdr, "Content-Type: %s\r\n", zContentType); | |
| 540 | + blob_appendf(&hdr, "Content-Type: %s%s\r\n", zContentType, | |
| 541 | + content_type_charset(zContentType)); | |
| 499 | 542 | if( fossil_strcmp(zContentType,"application/x-fossil")==0 ){ |
| 500 | 543 | cgi_combine_header_and_body(); |
| 501 | 544 | blob_compress(&cgiContent[0], &cgiContent[0]); |
| 502 | 545 | } |
| 503 | 546 | |
| 504 | 547 |
| --- src/cgi.c | |
| +++ src/cgi.c | |
| @@ -330,13 +330,36 @@ | |
| 330 | ** Return true if the response should be sent with Content-Encoding: gzip. |
| 331 | */ |
| 332 | static int is_gzippable(void){ |
| 333 | if( g.fNoHttpCompress ) return 0; |
| 334 | if( strstr(PD("HTTP_ACCEPT_ENCODING", ""), "gzip")==0 ) return 0; |
| 335 | return strncmp(zContentType, "text/", 5)==0 |
| 336 | || sqlite3_strglob("application/*xml", zContentType)==0 |
| 337 | || sqlite3_strglob("application/*javascript", zContentType)==0; |
| 338 | } |
| 339 | |
| 340 | |
| 341 | /* |
| 342 | ** The following routines read or write content from/to the wire for |
| @@ -419,10 +442,29 @@ | |
| 419 | if( !g.httpUseSSL ){ |
| 420 | fflush(g.httpOut); |
| 421 | } |
| 422 | } |
| 423 | |
| 424 | |
| 425 | /* |
| 426 | ** Generate the reply to a web request. The output might be an |
| 427 | ** full HTTP response, or a CGI response, depending on how things have |
| 428 | ** be set up. |
| @@ -493,11 +535,12 @@ | |
| 493 | |
| 494 | /* Content intended for logged in users should only be cached in |
| 495 | ** the browser, not some shared location. |
| 496 | */ |
| 497 | if( iReplyStatus!=304 ) { |
| 498 | blob_appendf(&hdr, "Content-Type: %s\r\n", zContentType); |
| 499 | if( fossil_strcmp(zContentType,"application/x-fossil")==0 ){ |
| 500 | cgi_combine_header_and_body(); |
| 501 | blob_compress(&cgiContent[0], &cgiContent[0]); |
| 502 | } |
| 503 | |
| 504 |
| --- src/cgi.c | |
| +++ src/cgi.c | |
| @@ -330,13 +330,36 @@ | |
| 330 | ** Return true if the response should be sent with Content-Encoding: gzip. |
| 331 | */ |
| 332 | static int is_gzippable(void){ |
| 333 | if( g.fNoHttpCompress ) return 0; |
| 334 | if( strstr(PD("HTTP_ACCEPT_ENCODING", ""), "gzip")==0 ) return 0; |
| 335 | /* Maintenance note: this oddball structure is intended to make |
| 336 | ** adding new mimetypes to this list less of a performance hit than |
| 337 | ** doing a strcmp/glob over a growing set of compressible types. */ |
| 338 | switch(zContentType ? *zContentType : 0){ |
| 339 | case (int)'a': |
| 340 | if(0==strncmp("application/",zContentType,12)){ |
| 341 | const char * z = &zContentType[12]; |
| 342 | switch(*z){ |
| 343 | case (int)'j': |
| 344 | return fossil_strcmp("javascript", z)==0 |
| 345 | || fossil_strcmp("json", z)==0; |
| 346 | case (int)'w': return fossil_strcmp("wasm", z)==0; |
| 347 | case (int)'x': |
| 348 | return fossil_strcmp("x-tcl", z)==0 |
| 349 | || fossil_strcmp("x-tar", z)==0; |
| 350 | default: |
| 351 | return sqlite3_strglob("*xml", z)==0; |
| 352 | } |
| 353 | } |
| 354 | break; |
| 355 | case (int)'i': |
| 356 | return fossil_strcmp(zContentType, "image/svg+xml")==0; |
| 357 | case (int)'t': |
| 358 | return fossil_strncmp(zContentType, "text/", 5)==0; |
| 359 | } |
| 360 | return 0; |
| 361 | } |
| 362 | |
| 363 | |
| 364 | /* |
| 365 | ** The following routines read or write content from/to the wire for |
| @@ -419,10 +442,29 @@ | |
| 442 | if( !g.httpUseSSL ){ |
| 443 | fflush(g.httpOut); |
| 444 | } |
| 445 | } |
| 446 | |
| 447 | /* |
| 448 | ** Given a Content-Type value, returns a string suitable for appending |
| 449 | ** to the Content-Type header for adding (or not) the "; charset=..." |
| 450 | ** part. It returns an empty string for most types or if zContentType |
| 451 | ** is NULL. |
| 452 | ** |
| 453 | ** See forum post f60dece061c364d1 for the discussions which lead to |
| 454 | ** this. Previously we always appended the charset, but WASM loaders |
| 455 | ** are pedantic and refuse to load any responses which have a |
| 456 | ** charset. Also, adding a charset is not strictly appropriate for |
| 457 | ** most types (and not required for many others which may ostensibly |
| 458 | ** benefit from one, as detailed in that forum post). |
| 459 | */ |
| 460 | static const char * content_type_charset(const char *zContentType){ |
| 461 | if(zContentType!=0){ |
| 462 | if(0==strncmp(zContentType,"text/",5)) return "; charset=utf-8"; |
| 463 | } |
| 464 | return ""; |
| 465 | } |
| 466 | |
| 467 | /* |
| 468 | ** Generate the reply to a web request. The output might be an |
| 469 | ** full HTTP response, or a CGI response, depending on how things have |
| 470 | ** be set up. |
| @@ -493,11 +535,12 @@ | |
| 535 | |
| 536 | /* Content intended for logged in users should only be cached in |
| 537 | ** the browser, not some shared location. |
| 538 | */ |
| 539 | if( iReplyStatus!=304 ) { |
| 540 | blob_appendf(&hdr, "Content-Type: %s%s\r\n", zContentType, |
| 541 | content_type_charset(zContentType)); |
| 542 | if( fossil_strcmp(zContentType,"application/x-fossil")==0 ){ |
| 543 | cgi_combine_header_and_body(); |
| 544 | blob_compress(&cgiContent[0], &cgiContent[0]); |
| 545 | } |
| 546 | |
| 547 |